Last night, I was super, super happy to invite my two good friends and idols, Hai Xin and A Wen, to do a live stream with us.
They came to share the full-process creation of their AI video work, "Louvre Cat."
I was stunned after listening. I said, you guys really tell everything, no holding back at all, just sharing it all like this...
Seriously, this live stream was packed with so much practical info. I didn't even want to summarize it because there was just too much. I spent the whole night organizing this transcript in detail, matching it with their presentation PPT, and now, with their authorization, I'm sharing it with everyone.
I hope these valuable experiences can provide some inspiration for our use of AI and our creative work.
After all, there aren't many people as talented as them who are willing to come out and share everything without reservation.
Everything starts with the film. This is the starting point.
Ring Hyacinth
@ring_hyacinth
·
At the invitation of the Pudong Art Museum, we created an official promotional video for the Louvre's first exhibition in Shanghai using AI. Hope you like it!
Louvre Museum Shanghai Debut: "Miracles of Patterns: Masterpieces of Indian, Iranian, and Ottoman Art from the Louvre"
December 13, 2025, to May 6, 2026, a major landing at the Pudong Art Museum.
Host: Louvre Museum, Pudong Art Museum
Producer: Lujiazui Group
Show more
73 162 992 135K
These two films were truly breathtaking to me at the time.
Art, this is true art.
This live stream is a no-holds-barred sharing by the two main creators of these two films, Hai Xin and A Wen, regarding "Louvre Cat."
The following is the live transcript, all in the voices of Hai Xin & A Wen:
————
Hello everyone, we are Hai Xin and A Wen. We are digital artists and a creative duo working with AI.
Today, we want to share the entire process—from concept to storyboarding to AI tool execution—of the official promotional videos we recently produced for the Pudong Art Museum.
We previously made two relatively famous short films. The first was early last year, when Beijing Daily invited us to create a city civilization promo for Beijing titled "The Forbidden City Cat Goes to Work." After completion, it spread very well on the internet and brought many opportunities, including an exhibition at the Osaka Expo and being the opening film for Unit 9 at the TED 2025 main venue.
During that TED event, Sam Altman also did an interview in Unit 11, so we were honored to share the same stage for a short time. We never imagined our films would travel to countries we haven't even visited ourselves.
The second film was this year's city civilization promo for Pudong, Shanghai, continuing the "kitten" theme, titled "Civilized Kitten Tours Pudong." It caused quite a stir upon release, with reposts from official accounts like Shanghai Release and Pudong Civilization, and the data was excellent.
Offline, it appeared on the ultra-long screens in the Lujiazui subway station underpass and the super large screen at Super Brand Mall facing the Oriental Pearl Tower, rotating in the core area of Lujiazui. It also played on the 12-screen array under the Oriental Pearl. We even received an "Internet High-Quality Communication Work Award" from the Shanghai Municipal Cyberspace Administration. We've always had a series related to kittens; you could say we are professionals at making cats.
At the end of the year, the Pudong Art Museum approached us. They were hosting a major official Louvre exhibition in Shanghai—the first large-scale exhibition of its kind in the city—and wanted us to create the official promo. The museum wanted two short films to cover a long promotion cycle of about one to two months.
We ultimately delivered two films. The first half tells the story of a "white kitten from France coming to Shanghai," while planting Easter eggs for the second half.
The second half tells the story of an "orange cat spokesperson for Pudong going to the Pudong Art Museum to see the exhibition."
Both films revolve around actual exhibits for narrative design. The first half centers on the Peacock Pattern Plate. The second half centers on the "Poetry Competition Tablet." In the second half, we also reprocessed the music, adding saxophone for a jazz variation to give it more of a Shanghai flavor.
After the films were released, they spread across various platforms, and the museum's official video account data was great. Our own accounts received many private messages from viewers saying they decided to see the exhibition because of these two short films.
There are also offline playback scenarios, such as the Bund screens, which will loop until next May.
Next, we want to share our full creative process and some insights, hoping they will be useful to you.
I. Casting
The first point we want to share is casting, and why we ultimately chose a white cat plus an orange cat.
Although we decided to continue with the cat theme, there's a lot of thought behind the cats themselves. Initially, we started from the theme colors of the Pudong Art Museum. The museum is primarily black and white, so our first thought for a protagonist was a cow cat (black and white).
The story in the initial version was completely different from the final film. The core exhibits weren't the Peacock Plate and the Poetry Tablet; we chose an oil painting. We set it so the cow cat was originally sleeping at the king's feet in the painting, then got "shaken off" when museum staff moved the painting. The cat begins exploring the Louvre and finally sees the exhibits leaving the Louvre in a truck bound for Shanghai. The cow cat decides to hop on a skateboard to chase the truck, with many thrilling small events happening along the way. We even made a demo.
We later abandoned this plan for several reasons. The most direct one was the long promotion period; the museum wanted two films. If we did the truck chase, the second part would look like a road movie after being split, which would skew the tone and consume too much energy on the chase, inconsistent with the vibe of an "official museum promo." So we scrapped the cow cat plan entirely.
After the museum requested two films, our first intuitive solution was simple: split the cow cat into a white cat and a black cat. Moving forward, we found that a black cat would appear too dark in the final film and wouldn't catch the eye. We wanted the protagonist to be "brighter" to grab the audience's attention faster, so we adjusted from "one black, one white" to "one white, one orange." Also, the orange cat echoes the orange cat's role as a Shanghai spokesperson from our previous film, making it more suitable for the Shanghai part. The French part was better suited for the white cat.
II. Setting the Tone
The second point is setting the tone.
Decide on the visuals and music first, then move forward with all the storyboards.
When making a film, a very necessary step is deciding the "film tone." Tone consists of two parts. The first is what the core visuals look like and the feel of the cinematography. The second is the music. For a TVC, music directly tells the audience how to feel and conversely guides the editing rhythm, determining whether shots are fast or slow cuts.
This project involved a lot of information: Pudong Art Museum, the Louvre, Islamic art, Shanghai, Paris, etc.
We first noticed the "Mirror Hall" of the Pudong Art Museum and felt the concept of mirrors was perfect for the structure. The Louvre and Pudong Art Museum could have a mirror relationship, as could Shanghai and Paris, and the two kittens. So we initially wanted to do a "split-screen animation," telling Paris on one side and Shanghai on the other, finally meeting at the Pudong Art Museum.
We did some early exploratory visuals, like the Bund looking at the Pudong Art Museum on top and the Louvre on the bottom. We also did a composition of two cats looking at posters on their respective desks.
Later, A Wen did a "radical overhaul" based on these visuals, placing the "Louvre" on top and the "Pudong Art Museum" as a reflection below. The image was very airy, with a cinematic texture, elegant temperament, and realistic, bright lighting. Another set of visuals was optimized, moving the "poster" from the desk to the building itself to make the environmental information more natural.
Based on this, we determined the visual atmosphere of the entire film.
III. Music
The third point is music.
We generally do music very early because it determines the rhythm, which determines the number of shots and the editing strategy. For this project, we decided the main instrument would be the piano. There are two reasons. One is that we like the "water ripple" reflection texture in the mirrored visuals, which reminds us of clean piano chords. The other is that the grid structure of the Mirror Hall reminds us of the order of a piano keyboard.
Narrative short film scores need an "arc"; they can't have the same rhythm from beginning to end like white noise. A story must have a beginning, development, climax, and conclusion, and the music must have a corresponding structure.
We used Suno because it generates good-sounding music and allows for fine control over segments. You can specify the length of each melody and clearly tell it what emotion to evoke at a certain point, such as a sudden turn, tension, or suspense.
Our music structure this time was roughly: the opening uses very simple piano chords to slowly draw the audience into the story. Then a segment follows the protagonist and the environment. After that, there's a transition segment creating a small setback to keep the audience watching. Then it enters the second melody, carrying the protagonist's actions and the climax. Finally, the ending—we wanted to return to a beautiful piano chord, so we specifically added an outro.
One advantage of Suno is that you can infinitely regenerate segments you're unhappy with. After finishing the French music, we exported the whole song and uploaded it back to Suno to make a variation for the Shanghai part while keeping the melody. We set the Audio Influence to 50%, ensuring the basic melody remained, then told it to add saxophone and make it feel like old Shanghai jazz. It made very appropriate variations on the original melody.
It once gave a rather exaggerated ending, but we eventually changed it back to the original ending to keep it more restrained.
Additionally, we post these scores online under the identity of our cat, Nika. Nika is considered a "legendary musician on the Cat Planet," and it includes some TVC scores we've done for clients.
IV. Storyboarding
Next is how to approach storyboarding.
Our personal aesthetics are very strong, so this is more about sharing experience.
Our basic requirement for the opening storyboard is "high information density."
Try to hint at what the story is about in the very first shot. For example, the first shot of "Forbidden City Cat Goes to Work" is the cat opening its eyes with the reflection of the Forbidden City in its pupils, quickly explaining "cat and Forbidden City." The start of "Civilized Kitten Tours Pudong" is the cat pushing open a map of Shanghai, quickly explaining "cat coming to Shanghai for tourism."
For the Louvre white kitten part, we wanted to explain the story within three shots. The first shot has the cat looking at the poster, planting the "exhibition poster" key visual in the audience's mind. The second shot uses a close-up to introduce the protagonist. The third shot uses a wide shot to establish the environment while showing the mirror relationship of "Louvre reflecting Pudong Art Museum."
Storyboarding also needs a sense of rhythm. Rhythm largely comes from changes in shot size. Wide shot after wide shot feels very monotonous, so we cut back and forth between wide shots and close-ups to create a sense of expansion and contraction. If the opening shot has too much information, the audience won't know where to look in the first second; we use simple animation to guide the eye. The first half uses a "pulling back the curtain" method, and the second half also uses a classic opening design.
Then comes the first part of the story, which is a core part and a set of storyboards we are very satisfied with.
The purpose of this sequence is to show the white cat walking out of the Louvre, falling in love with the Peacock Pattern Plate, and having a fantastic imagination.
This sequence needs to convey two pieces of information: the protagonist is at the Louvre, and the protagonist likes the peacock plate.
The first shot uses a classic Louvre exhibit to establish the location. We chose the Winged Victory of Samothrace, a very classic sculpture. We deliberately didn't choose the Mona Lisa because it's too common and would seem too cliché. In terms of composition, we didn't focus the camera solely on the sculpture because we wanted the audience to focus on the cat. So we placed the sculpture in the background as an embellishment and used an extreme low-angle shot, making the white cat in the foreground occupy more volume in the frame, forcing the audience to focus on the cat while still instantly recognizing "this is the Louvre."
The second shot begins to subtly introduce Islamic art. We referenced an artist famous on Instagram whose common practice is to film daily life scenes and then make the symbols within them move.
That feeling is wonderful because it's fantastical yet feels like something that could happen in real life. We initially made a more exaggerated version, like the whole scene turning into patterns as the cat slides by, but later felt it was too much movement and didn't feel like real life, so we changed it to a more restrained fixed-camera plan, letting the background patterns move subtly as the cat walks by.
We tried many versions: magic-circle-like patterns, the camera panning down into an Islamic art world, patterns protruding from the screen, etc. Later we realized that "interesting" wasn't enough; we also needed emotional gain. So we used a contrast design: starting with an oppressive extreme high-angle shot, and as the camera moves to the end frame, it enters a huge open space, giving the audience a sense of sudden enlightenment. In the final shot, the patterns on both sides move like a conveyor belt, bringing the cat into an open exhibition space.
We also made a trade-off: no portraits in the frames. Once a human portrait appears in the frame, the audience will involuntarily look for all the portraits, and their attention will be stolen from the cat. We wanted the audience to follow the cat, so we removed elements that easily steal the show as much as possible to keep the expression subtle.
In the end-frame design of this shot, A Wen performed a very clever operation: placing the Peacock Pattern Plate directly into the exhibition space, letting the core exhibit appear early during the climax. After determining the end frame, we went back and redesigned the first frame and camera movement, making the shot more controlled.
After the emotional climax, we needed to add a bit of new stimulation. We thought about the cat seeing the space after going up, or seeing another exhibit, but none were touching enough.
Later we thought of the cat seeing a peacock, and the peacock is alive. This point immediately made us feel like the artwork had come to life. In terms of rhythm, we also did a step-by-step progression: the peacock first turns its head, then its body, and finally spreads its wings, saving the surprise for the last second.
After the climax, it needs to return to reality. We used a very simple way: cutting back to a close-up of the cat, as if it's imagining, then cutting to an objective shot establishing reality—the cat is actually standing in front of the peacock plate looking at the exhibit. The audience then understands that the fantasy segment came from the cat's imagination.
Then comes the transition segment, aiming to explain that "the exhibit is going to Shanghai" and "the cat decides to get in the box and go along."
We used montage and split-screen processing because it can convey a lot of information in a short time. When making such shots, we cut out the people as much as possible. In animal films, human faces easily steal the show; audiences are more likely to empathize with people, and their attention will be taken away, so we only kept close-ups and actions without showing faces.
The second segment was originally going to be about what happens to the cat in the airplane cargo box.
Before fully determining the plot, we use Sora to run some samples to quickly check rhythm and composition, looking for surprises and references. After running them, we didn't like it because the images weren't airy, the colors were yellowish, old, and dark, inconsistent with the tone we set earlier. The plot logic also didn't work because the cat in the box couldn't see the exhibits outside, and the exhibits should also be in boxes.
So we scrapped the entire "peeking inside the box" segment and pulled the focus back to the core narrative: the white cat accompanying the peacock plate to Shanghai.
For the arrival in Shanghai, we initially wanted to do a realistic transition of a plane flying from the Louvre to Shanghai, and we ran it with VEO 3. We also tried a classical map micro-animation. But these plans were all rejected by us.
Because a map makes the audience particularly concerned about rationality, such as whether the positions of Paris and Shanghai and the flight distance are correct, which distracts attention. We also tried a more abstract "fire point map with footprints" plan, but we didn't like the shot being too wide; we preferred the audience to see the key subject occupying a larger portion of the frame.
Finally, we decided to change directly between the "Louvre first frame" and the "Pudong Art Museum first frame," making the transition more close-up and focused.
We tested many types of transition animations: like a carpet unfolding, tiles flipping, airplane imagery, etc. We finally chose the airplane because the airplane imagery appeared in the shots before and after the transition. Continuous imagery makes the audience feel more comfortable, even if it jumps from realistic to a mosaic art style in the middle, it won't be abrupt.
After arriving in Shanghai, the storyboarding is more direct: first seeing the Pudong Art Museum from far to near, then seeing the cat running toward the museum. We really like one of the shots: the cat running by the water, with the reflection in the water being a peacock.
This idea was initially "wilder." We hoped that as the cat ran, it could see the shadows of many animals, like peacocks, horses, camels, etc., bringing out a sense that "all Islamic art has arrived in Shanghai together." We tried text-to-image and text-to-video, but none worked. We also thought about the cat seeing itself as a tiger in its heart, with the shadow being a tiger, but this was too weakly related to the film and was quickly removed.
We later found that handing such core creativity directly to AI in one go is unreliable and needs to be broken down for execution. We first hand-drew the storyboard, deciding on a slightly high-angle shot so the audience's attention naturally falls on the lake reflection.
The composition was simplified to "the cat's shadow is a peacock," which is most directly related to the main line. We would first use Photoshop to composite the desired relationship, then hand it over to the model to complete the unified style and dynamics. Throughout the process, we strongly felt that even a crude hand-drawing is more intuitive for AI to understand and for people to communicate.
For the closing, we echoed the concept of "mirrors" again, ending with a more restrained and beautiful image.
That's all for the storyboarding.
Next, the art part is handed over to A Wen.
V. Art
I (A Wen) mainly do two things.
First, convert Hai Xin's manuscript storyboards into final visuals. Second, control the overall art style.
After receiving the project, we immediately went to learn about Islamic art. It's an art style we usually have little contact with, but it's very distinctive. After mastering some core keywords, it's easy to create mosaic-style art with exotic features. We summarized 4 very useful keywords in the live stream; the two most commonly used are "Iznik style" and "Mosaic art." Here, mosaic refers to Mosaic art, which is different from Pixel art.
We have two principles for making films. It must look good at first glance and be durable upon second glance. In the AI era, it's a precious thing for an audience to be willing to finish watching your short film. Looking good to us equals a cinematic feel. Durable equals as many details and creative details as possible, including art details.
Everyone is actually very sensitive to a cinematic feel. For the same image and the same prompt, with results from different models, you can easily judge which one has more of a cinematic feel. So the first step in art is choosing the right model. Our current main image generation model is basically Nano Banana Pro. For image modification, we mostly use it directly with text modification, sometimes combined with Photoshop. I also manually fine-tune some parts.
Choosing the right model is as valuable as choosing the right vendor. It can do the job well and handle many grueling modifications. For example, there was a shot we were very satisfied with, but after submitting the first draft, the museum said this shot had to be removed because it would make moving the exhibits look unprofessional, and they didn't want to convey that feeling to the audience. Narratively, we needed this shot to explain the cat hiding in the box to follow the exhibits to Shanghai. Finally, we used Nano Banana to "wrap" the exhibits, making the handling look more professional, and it passed.
Another example: the last frame of the image is very beautiful, but you'll find that attention stays on the white cat's butt for a long time; this was a problem exposed only after switching to the white cat. The solution was simple: let the tail hang down in the first frame, and it wouldn't be an issue when walking up later.
Even more exaggerated was saving the day by changing the cat. When we switched the protagonist from a black cat to a white cat halfway through, many static frames and animations were already finished.
It sounds outrageous, but AI can really change it.
At that time, the tools we could use were limited, only the first generation of Nano Banana, which we usually used in Google's Whisk. Our habit is one folder per shot, so we went into each folder one by one to change the cat's color, and finished it all in a day and a half.
Later, when reviewing, Banana Pro came out, and we found that even more outrageous efficiency was yet to come. AI can not only change it but also change it faster and faster. On third-party platforms, you can upload all static frames at once and change the black cat protagonist to a white cat with just one sentence, which is more efficient.
Many people ask where Nano Banana Pro is best used.
Our experience is that Google's own membership system has quite a few pitfalls. We've bought the top-tier Google Ultra membership since it launched and have used it until now. You often see others saying Google has discounts, like $79 per month for three months, or a half-year price, but looking back at your own subscription price, it might not change at all and remains very expensive, even higher with tax.
Another point that makes us very unhappy is that the 4K entry point is unclear. Some say it's in Flow, but Flow's default generation is 1K, and it only super-resolves to 4K when downloading. We're not sure if that's super-resolution or native 4K. Google AI Studio can output 4K more natively, but it's charged separately.
If you're an Ultra member like us, here's our more pragmatic usage: use Flow for quick gacha-style generation because it's fast with a 1K base. Use Gemini for multi-round iteration; the dialogue rounds are very convenient. When you have a good grasp of the result, you can go to AI Studio for native 4K.
Additionally, our prompts are getting simpler and simpler. The model's multimodal understanding is very strong, and often doesn't require long essays. For example, I'll throw a picture in, let it extract the cat and butterfly, put them on a solid color background, and then turn it into pixel art. After stabilizing the style, extending the materials is very efficient.
And do today's prompts really need to be long essays?
We had a shot that needed a front view of an exhibit, but no one could get in during the setup period; the staff could only send us two quick snapshots from afar.
These two were already very precious to us, so we used them as intelligence and directly let Gemini generate a front view of the exhibit based on the two photos, in 16:9.
The result was very good, and we used it to complete the final shot.
Even more surprising was when we went to the site on the opening day and found it was almost identical to the actual location. People couldn't squeeze in, but the AI did.
When a model is powerful enough, it doesn't care about the origin of the storyboard. That is to say, sometimes you give it some rather outrageous storyboards, and it can instantly get what you want and directly generate some very nice visuals. These are all visuals directly output from the model using Hai Xin's storyboards, and they are all very good.
But for complex creativity, you must iterate in stages.
Take the shot of "the cat walking on the shore, with the reflection in the water being a peacock" as an example.
First, I would understand the perspective of the storyboard sketch.
I'm used to building a very rough sketch in PPT. I happened to have a transparent PNG of a white cat "veteran actor" on hand; putting it in the frame builds the perspective. Then I throw this rough sketch into the model to turn it into a realistic style, first getting a basic scene of "a cat walking on the shore with a lake nearby."
The second step is laying the floor tiles. I indeed used prompts to lay floor tiles on the concrete ground, trying many keywords until I found a version with the "just right density." After laying them, the image took shape, but multiple iterations would deform the cat's posture, so I'd let the AI adjust it again.
The third step is adding exhibit information. The simplest way is to overlay it in PS and adjust the layer style. But we found that the peacock reflection in the lake was hard to get moving because stylized peacocks aren't easily converted into realistic dynamics. So I went back to PS, erased the peacock, replaced it with a realistic peacock shadow, and threw it back into the model to continue running.
Later, the cat's color had to be changed. After changing it, I found the floor tiles "ate" the white cat, so I relaid the floor tiles again to get the final static frame.
Finally, making the animation was actually the simplest part. Usually, it's direct output; two or three tries can get good dynamics, like the cat walking along the shore and the peacock reflection in the water following along.
Then some experience in art regarding real-scene compositing.
There are many landmarks and exhibits in the short film that cannot be wrong, so we need certainty. Many times we composite the cat onto real existing landmarks.
Now this kind of compositing is very simple; you just tell Nano Banana Pro that there's an orange cat with its back to the camera chasing a butterfly of a specified color in the scene, and it can generate it.
Two small tips to improve the success rate. First, crop before generating. For example, we didn't need the decorations at the bottom of the Peace Hotel material, so I'd crop them out before giving it to the model. Many indoor shots are the same; crop out unnecessary areas first, and the model becomes more focused.
Second, if you need to do a match cut or repeatedly reuse a subject, it's recommended to first extract the cat and butterfly onto a solid color background and then change various backgrounds. Letting the model constantly change scenes directly in the same image easily eats the butterfly, eats the cat, or changes the cat's color. Extracting the subject first and then changing the background has a much higher success rate.
Many people ask why we don't use multi-image reference for video generation.
Regarding multi-reference video products, we tried almost everything on the market while making these two short films, and the clarity issue is hard to cure; elements become blurry with even a slight movement.
Even with the new multi-reference function recently launched on Flow, we tried it, and it's hard to ensure the exhibits we want remain unchanged. We have too many stable subjects to maintain, including multiple exhibits, the cat, and the butterfly, so in the end, the first-and-last-frame routine is more stable.
Many friends also ask, is there still a need for PS?
I think PS is still useful, but we open it less and less frequently. I highly recommend a function that not many people use, called "Harmonize."
You paste a PNG or image onto a static frame, click harmonize, and it automatically matches the environmental lighting, making the composite more integrated. For example, in the shot with the floating plate, I click harmonize, and the difference in lighting before and after is very obvious; the efficiency is high.
At the same time, Banana Pro is powerful enough that many things can be done directly with it. For example, in the orange cat film, there's a shot where I first used Vidu to generate a preview; the clarity wasn't enough, and the exhibits weren't stable, but the relationship between the cat's movement and the exhibits was right.
So I threw both the preview and the exhibit materials into Banana Pro, letting it keep the composition unchanged and replace the floating exhibits with these I provided, allowing them to appear repeatedly and in a scattered manner. The result was a very good image in the first version, and it even automatically added foreground/background depth of field and motion blur, basically meeting production needs.
VI. Animation
Our main video tool is Flow. Our commonly used video model is VEO 3.1.
Flow also recently launched a 4K function, which helps with image quality. For auxiliary video models, we also use Kling, Hailuo, Jimeng, Wanxiang, Luma, etc., but Flow is more convenient to use, and the visuals have more of a cinematic feel.
We feel that we have entered an era where we can "edit video." Many video tools can directly edit video. There's a hidden function in Flow, with a deep entry point; there's an "edit" button in the top left of the generated video. Once inside, you can add things to the video, cut things out, and even change the camera position and movement; it's a bit experimental but sometimes useful.
Its limitations are also obvious; for example, it can add and cut, but it's hard to do "edit and replace," like changing a black cat to a white cat. To achieve this, you have to remove first and then insert, which feels very clunky.
In Flow, we also use the doodle feature a lot. Pure text is hard to get the cat to complete certain actions, like having the cat jump directly across; we couldn't get it even after many tries.
But by drawing a doodle on the first frame to give movement instructions, like where the butterfly should fly and the cat following it, and then adding a prompt "follow the instructions in the annotation and then delete my annotation," the model will generate according to the movement rules you annotated, and the cat can smoothly walk out of the frame.
And Luma's Ray3 is unexpectedly useful. We did a test at Christmas, uploading a very rough preview and adding a prompt; it generated very beautifully with high clarity, even creating something out of nothing in the original video.
The disadvantage is slow generation, but the advantage is high compositing quality, suitable for production.
Wan 2.6's new character creation feature is also worth noting. Its human characters are still prone to being out of character, but animal characters are much more exciting than humans. I casually uploaded a very blurry and dark video of my cat Nika, and after building the character, I could use @ to summon it in Wan 2.6; the generation quality was at a "master-approved" level and very high-definition. We had it perform many things, like a dinosaur coming to save it. This feature launched after our two short films were already posted, which was a bit of bad timing.
Another small tip: when making micro-animations for the end screen, you can choose a "slightly worse model," i.e., a previous generation model. Previous generation movement amplitudes are usually smaller, which is sometimes just right for the end screen. For example, for the end screen, we want the environment to move slightly, but Flow might add many imagined patterns and move too much.
Using a previous generation model actually moves just right. I used Jimeng 3.0 this time, and that micro-movement amplitude was very appropriate.
VII. Discarded Drafts
There were many discarded drafts this time, especially for the orange cat one, because after Pro came out, implementing some ideas was fast, and the amount of exploration was greater. Here are a few we particularly liked but painfully deleted.
One is the "Butterfly Pastry" that Shanghai audiences love to ask about. Shanghai's butterfly pastry is very famous, and we indeed made a giant butterfly pastry, and we liked the visual very much. But in AI, once the butterfly pastry moves, it turns back into a real butterfly and never flies, so we had to give it up.
Another is a very popular check-in spot at the Pudong Art Museum where you can photograph the Oriental Pearl. We made a shot where a butterfly sticks half a wing in from outside the window, and the other half of the wing is completed through the shadow of the sunlight, forming a complete butterfly. The visual was beautiful, but when it moved, the butterfly still wouldn't fly, so it was deleted.
Another was a shot of the Louvre's Islamic art gallery; we initially wanted to make it more exaggerated, even with tsunami-level dynamics, but later felt it was too fantastical and didn't fit the "low-key but fantastical" feel we wanted, so it was also deleted.
Finally
Finally, the core point we want to convey is:
The more convenient the model, the more you should push yourself to do better.
Now that models are getting more and more convenient, many creators tend to use agents to directly output entire sets of storyboard solutions or let them complete more of the creation. But what we want to say is that model convenience should not be an excuse for laziness.
It's more like a reminder: you have more ability to make your work better.
Ending with a phrase we often say:
As long as you are in action, you are moving forward.
Thank you everyone, our sharing for today ends here.
If you found it useful, feel free to forward it to your friends to take a look; it's a huge help to us!
