ComfyUI Tutorial Series 61-70

I tested this animation in the previous lesspn 48 LTX 0.9.7
In LTX, it is possible to add two frames between which the animation takes place.
In Wan2.2, this version also has the ability to add a start and end image for the animation.
This version of Wan2.2 generates animations better, more detailed than LTX 0.9.7. But there are errors in the animation.
For example, a meteorite can fall slowly, can fly from different directions, even flew from the ground once, if you do not specify where the meteorite is flying from. The speed of the meteorite is not set, even if you specify that it falls quickly.
When a meteorite falls into a car, incorrect animations may occur. The meteorite may fall, lie down, start to rise up, then smoke will come out or a second hood may rise from the car, the first hood of the car does not move.

I did this animation in parts: first the camera approaches, then the door opens, the hare appears, then he comes out and says hello. Here I didn't have a frame in which the door is open, so during the animation the door should disappear, because when the hare appears there is no door. I made an animation of the door opening and added the door to the frame when the hare appears.
There are small inconsistencies with the speed of the camera movement, because when approaching the speed is one, then in the next animation the camera approaches at a different speed and this transition is not smooth. I edited these videos in parts.
I made an upscale video, but when upscaled, the video is not detailed by the prompt and is not corrected if there are errors, as was done in ltx.
In the next lesson 69 there will be LongCat video, there is an opportunity to create a whole chain of prompts for one image and set complex scenarios for animation. This allows you you not to edit video pieces in parts.

I modeled this room in Cinema 4d lesson, Corona render.
I tested the motion animation in Wan2.2 on frames that I modeled in Cinema4d.
The camera motion animation is almost correct, except for the glitches in the movement of the beams on the top right, because the AI ​​is confused about which beams should move where. You can test other generations, add clarifications in the prompt and test how to fix it. Such an animation should be run correctly in Cinema4d, and there will be a higher quality correct animation with the detail of the tree and other objects, but it will take more time to generate all the frames.
I added a prompt in this animation for people to gesture, but because the people are small, the AI ​​does not animate them.

I generated animation in Wan2.2 horizontal video with an enlarged angle of people.
I made a camera zoom animation (Image scaling) and added a generated animation with people gesturing.
At a distance, people are poorly recognized by AI and do not animate, so they are initially static.

Here the author shows how to generate in the cloud Image2Image on Nano Banana

Anime animations on the NetaYume v3.5 model are generated with sounds, and melodies are added to the background.
But sometimes melodies get in the way and are added even when the prompt says that melodies are not needed.
Here the author shows how to generate in the cloud Image2Video на Sora2, there is a link from midjorney-plus.oss-west-1.aliyuncs.com/sora to an mp4 that is downloaded without compression, this link can be downloaded, then it is deleted there.

I tested in in the cloud on the Sora2 Image to Video model, the video is not fully animated, this is a complex prompt for this image. The hare does not walk, but loops, perhaps it is worth testing this generation model more.
Sparks fly well in the animation, sounds are added in the forest, birds sing.
In previous lessons I tested a lot of animation generation on different models with this hare.

This is an animation of a hare in the forest sniffing mushrooms on the LongCat model.
First, the hare had a mushroom in its paw, I made a more detailed prompt, that the hare approaches the mushroom, stops, rejoices.
In the animation, the hare is not just rejoiced, he takes the mushroom, although this is not indicated in the prompt, you can further test the prompt.
The animation of the hare's movement is high-quality, although in the second video the mushrooms in the foreground rise slightly during the hare's movement.

Animations with a girl sucking chupachups rarely turned out correctly. It is necessary to clarify prompt, that a girl puts a lollipop in her mouth, licks it. It was rarely generated correctly. Generated once clearly correct, but there was a glitch in the tongue movement. There were many animations where the girl doesn't lick the lollipop, but just moves her hand with it.

The animations of the mouse from the house and its movements on the LongCat model are correct. The first animation is without glitches.
There are some flaws: sometimes the door may not open correctly or disappear, the light may turn on but not be reflected in the windows, the door opening may wobble. You can specify the prompt in more detail to get better animation.

In total, there were 6 videos of 91 frames in the 768x768 video, resulting in 31 seconds.
Prompts for each part:
1. A young woman with curly purple hair sits at a cozy café table, wearing a black turtleneck. The camera is static, centered in front of her. She gently stirs her coffee in a white mug with a spoon, soft warm lighting and a calm atmosphere.
2. the woman lifts the spoon from the mug and sets it on the table, Her movements are natural
3. the woman picks up the mug with both hands and takes a slow sip of coffee
4. the woman place the cup on the table and smiles to the camera
5. a black cat walking into frame from the side, jumping onto the table next to the mug. The woman looks at the cat with surprise and a gentle smile.
6. the woman take her hand from the mug and reaches out to pet the cat, smiling warmly as the cat relaxes beside her

The recommended size is maximum 768x768.
The LongCat model now has the ability to make many prompts for the following parts of the video using one image.
This allows you not to edit them separately. At each stage of video generation, you can test a separate part of the animation and move on to the next one if it suits you. Video generation requires a very powerful video card, you can generate it on the site runninghun.ai via the author's forkflow link LongCat img2video - 5 to 30 sec.
There are some minor glitches in the generation that are fixed by tweaking the prompt and generating other variants. For example, in one part of the animation a black cat appears, then another cat appears, without specifying that the cat is black.
Maximum generation sizes: 720x1280, 1280x720, 768x768, no 1024x1024 size.
I tried to generate 1024x1024 on a test account, necessary to do this on a paid account, it might generate at 1024x1024, but I read that the maximum square size for LongCat in this version is 768x768.

Prompt for video:
Slow cinematic camera tracking forward, low angle, following a cute fluffy white bunny with big blue eyes and rosy cheeks as it walks joyfully along a winding path in an enchanted forest. The bunny wears a small brown backpack, its soft fur bouncing with each step. Surrounded by oversized glowing mushrooms casting warm light, colorful bioluminescent flowers, and floating magical sparkles, the scene feels whimsical and dreamlike. Tall, curved trees arch overhead, with gentle rays of sunlight filtering through the leaves, creating a serene, fairy tale atmosphere.
The animation on LongCat is high-quality, the movement of the hare is correct, light. Sparks are animated correctly. The hare did not lose detail during the animation, for example, on wool, the camera has depth of field, there is focus on the hare and detailed grass is visible, and in the near background the grass is blurred. Good lighting with the sun's rays, as in the picture. Does not add sounds as it did model Image2Video Sora2 in the previous lesson 68.
The only thing is, at the end, the camera can be moved a little so that it doesn't cut off the hare, you can test other options. Overall, the animation is good, already approaching the quality of paid models. In the latest lessons I have extensively tested animation generation on various models with this hare.