ComfyUI Tutorial Series 51-60

Prompt for video:
adorable white cartoon bunny with big eyes and pink cheeks walking happily down a sunlit forest path, surrounded by tall trees, soft grass, and blooming daisies, warm morning light filtering through the leaves, cinematic depth of field, vibrant colors, soft shadows, slow cinematic camera movement

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.

I generated this girlin the episode 28.
Prompt for video:
A cute cartoon girl with purple hair walks slowly along a sunlit wooden path surrounded by sunflowers. She smiles warmly, her gait graceful and natural. The scene is bright, colorful, and cheerful. The camera gradually moves away and shows the girl, the girl always remains the same size in the frame. The girl walks forward towards camera, camera moves back. The girl's steps are correct, the girl does not slip, the girl does not limp, she walks a distance equal to her step size.

I generated this womanin the episode 28.
Prompt for video:
A beautiful young woman with short green hair walks gracefully through a sunlit forest path, wearing a flowing yellow dress with red floral embroidery on the chest. The sunlight filters softly through the tall trees, creating a warm, dreamy atmosphere. The camera smoothly tracks backward, maintaining the same distance as she walks forward with effortless elegance. She smiles slightly, her expression calm and confident. The fabric of her dress catches the light as she moves, and the background bokeh glows with golden tones. Cinematic, natural lighting, shallow depth of field, gentle camera motion, 4K, ultra-realistic.

Prompt for video:
360 degree camera rotation, slowly cinematic, woman smile subtly, white background

I generated this video in ComfyUI, workflow, on the model wan2.1_i2v_480_fp8_scaled and LoRA rotate_20_epochs.
This Lora was specially trained to rotate the camera around objects. The video is generated correctly, clearly one rotation of the camera 360 degrees around the girl, but the girl's smile is not visible. On some paid graphics generation sites, such a prompt to rotate the camera around an object will not generate the correct video without this Lora.
I generated this workflow in the cloud: runninghub.ai

Prompt for video:
360 degree camera rotation, slowly cinematic, woman smile subtly, white background

I generated this video on the website: krea.ai/video
I generated several times, but the camera did not rotate around the girl. In the second generation, the girl's body rotated, but her head did not move, although the prompt did not indicate that there was rotation, it generated wrong video.
In this video, there is a slight rotation of the girl by 45 degrees. Although the generation does not correspond to the prompt and the camera does not rotate 360 ​​around the girl, in my opinion the girl's smile turned out to be very lively and realistic.

Prompt for video:
Woman closeup, slow cinematic camera, she smile subtly, white background

I generated these two videos in which the girl is smiling in ComfyUI: workflow, on the model wan2.1_i2v_480_fp8_scaled.
The videos correspond to the prompt, only in the first video the girl does not smile for a long time. Before that, there were unsuccessful generations, a gray background was added instead of white. Only when I added in the prompt: "white background", the white background did not change.
The gray background was also added to the paid site in one generation even when the prompt said: "white background".
I generated these two videos in ComfyUI in the cloud: runninghub.ai

svg1    svg2    svg3    svg4    svg5    svg6    svg7    svg8    svg9    svg10    svg11    svg12    svg13    svg14    svg15

Prompt for video:
cute gnome sitting on a mushroom, moving with hand like it says hi

The animation matches the prompt. Sometimes the gnome waves with two hands instead of one.

Prompt for video:
cute gnome getting down from the mushroom and coming closer to the camera

At first I created a prompt that the gnome just jumps off the mushroom, but the gnome jumped in different directions not directly towards the camera. Added this to the prompt. The gnome jumps correctly, but does it for a long time, communicates a lot, lingers and does not jump quickly onto the grass. The gnome jumps like levitation or swimming in water. It is necessary to specify in the prompt the speed of his jump.

Prompt for video:
The gnome stand on a mushroom and jumps from the the mushroom to the grass

Some options are less successful and the dwarf can jump to the side or fly away like a bird, but sometimes the jump is correct and the dwarf bounces several times on the grass when he jumps. You need to test different options and edit the prompt to get a successful animation. You can add to the prompt so that the camera does not move and the dwarf does not jump on the grass.

Prompt for video:
sq41sh squish effect, there are five fingers on each hand

In this workflow I used in the lesson the model wan2.1-i2v-14b-480p-Q4_0.gguf (there are other versions, weigh more and are of better quality), model Lora: Phantom_Wan_14B_FusionX_Lora(improves video), Lora: squish_18 (trained model of squeezing various objects with fingers).
The animation is correct, the fingers correctly mimic the gnome. But sometimes there are six fingers on each hand.
Sometimes the gnome reacts to the movement of the hands, starts moving his hands, covering his eyes, as if he understands that he will be squeezed.

Prompt for video:
The gnome stands on a mushroom and dances

This workflowу I generated ComfyUI in the cloud RunningHub AI on models Wan2.2_i2v_high_noise_14B_Q5_K_S.gguf, Wan2.2_i2v_low_noise_14B_Q4_K_M.gguf, which were there.
The animation is correct, the gnome stands on the mushroom and dances. The mushroom leans in different directions under the gnome's weight.

Prompt for video:
a cute rubber duck toy floating on a pond, static camera

The animation is correct. You can do Upscale відео.

Prompt for video:
a cute bunny looking curious at the camera, adorable

The animation is correct. You can do Upscale відео.

Prompt for video:
Adorable young panda sitting in a wooden tub filled with colorful floating balls, nestled in a lush bamboo forest, soft golden daylight streaming through tall bamboo stalks, warm highlights on fur and water surface, natural depth of field emphasizing the panda’s expressive eyes. The panda looks directly at the camera with curiosity, gently splashing and playing with the balls in the tub. The panda throws the balls up. Joyful and innocent atmosphere, detailed textures in fur, wood, and bamboo. Camera fixed on a tripod in a steady medium shot, capturing the playful interaction without movement, cinematic framing with subtle focus shifts to enhance realism and storytelling.

The animation is correct. I did the last video Upscale below. Воркфлоу у ComfyUI у хмарі
In lesson 48 I generated such a video with a panda on a model Ltxv and did upscale, but you can see that it is not as detailed as on the wan2.2 model. For example, on Ltxv the color of the balls changed when the panda threw the balls, one ball became large, but it was not in the bathroom, the balls did not move on the water, the water splashes, although they flew correctly, were poorly detailed. On the wan2.2 video, all the balls float on the water separately, more realistic bubbles, water splashes, there are no camera failures. The workflow with wan2.2 understands the prompt better, the animation is more detailed, you can add other Lora models compatible with the Wan2.1 model, the video generation speed is longer than on ltxv, but the videos with failures are less.
Paid sites can generate this video better, but the free wan2.2 model generates normal videos that can match the prompt.

Prompt for video:
adorable white cartoon bunny with big eyes and pink cheeks walking happily down a sunlit forest path, surrounded by tall trees, soft grass, and blooming daisies, warm morning light filtering through the leaves, cinematic depth of field, vibrant colors, soft shadows, slow cinematic camera movement

The animation is correct. But the hare is walking slowly. Workflow in ComfyUI in the cloud RunningHub AI
In lesson 51 I generated such a video with a hare on a paid site hailuoai.video. t generates such animation better. But for some reason the hare turned and moved back from the camera, although other models generate the correct animation of the hare's movement at such a prompt. Also in lesson 51 I generated this video with a hare on a model Cosmos Predict2, which is free. You can see the similarity of the video Cosmos Predict2 і Wan2.2.

Prompt for video:
adorable white cartoon bunny with big eyes and pink cheeks walking happily down a sunlit forest path, surrounded by tall trees, soft grass, and blooming daisies, warm morning light filtering through the leaves, cinematic depth of field, vibrant colors, soft shadows, slow cinematic camera movement

The animation on Wan2.2 is correct from the first generation.
Image of a girl I generated in lesson 51 on the model hidream-i1-fast-Q8_0.gguf, workflow screenshot, when comparing the generation of different models with the model Cosmos Predict2.

I generated this girlin episode 28.
Prompt for video:
A cute cartoon girl with purple hair walks slowly along a sunlit wooden path surrounded by sunflowers. She smiles warmly, her gait graceful and natural. The scene is bright, colorful, and cheerful. The camera gradually moves away and shows the girl, the girl always remains the same size in the frame. The girl walks forward towards camera, camera moves back. The girl's steps are correct, the girl does not slip, the girl does not limp, she walks a distance equal to her step size.

This animation of the girl's walk on Wan2.2 is correct, I generated it on the first try. Wan2.2 began to understand the prompt better and generate videos better than Wan2.1.
In previous lessons I generated this girl on different models.
On Wan2.1 sometimes The girl moved back.
Often on Wan2.1 The girl's gait was irregular.
On Wan2.1 The girl's gait was erratic.
Although I managed to generate The correct animation of the girl's walk on Wan2.1, but there were many errors in the other videos.
I also generated High-quality animation of this girl's gait on the website openart.ai/video/i2v
I generated High-quality animation of this girl's gait on the website hailuoai.video

generated this woman in the episode 28.
Prompt for video:
A beautiful young woman with short green hair walks gracefully through a sunlit forest path, wearing a flowing yellow dress with red floral embroidery on the chest. The sunlight filters softly through the tall trees, creating a warm, dreamy atmosphere. The camera smoothly tracks backward, maintaining the same distance as she walks forward with effortless elegance. She smiles slightly, her expression calm and confident. The fabric of her dress catches the light as she moves, and the background bokeh glows with golden tones. Cinematic, natural lighting, shallow depth of field, gentle camera motion, 4K, ultra-realistic.

The animation on Wan2.2 of the girl's gait is correct from the first attempt. You can see that the face has hardly changed in the video. The colors in the video have not changed, but the camera has slightly changed the angle, adding more space from above. The girl's movement is correct, there is a smile. The head tilts slightly in different directions, the body and arms move correctly. The girl's hair also sways correctly with each step.
Previously I generated The correct 360 degree rotation animation of this woman with Lora additions on Wan2.1.
But there were many glitches with body parts on Wan2.1, incorrect movements. Tested other Lora with "With a kiss" and "Samurai" on Wan2.1.There were a lot of errors in video generation on Wan2.1: the animation was unclear because of what was indicated in the prompt, and the objects moved erratically. Wan2.2 began to understand the prompts better and animate better.
Previously I generated High-quality animations of this girl's gait on the website Runway. Runway saved the girl's face in the video with high quality and added a lively smile to her.
I generated Correct animation of this woman's gait on the website Hailuo AI. The woman is holding the hem with one hand, but the girl's face has changed and the contrast and brightness in the video have increased slightly. The girl's face in the video was generated better on Wan2.2 in this lesson.