The effect is not very good

[hegc]

Is there an instruction manual?

img(gsl.jpg):

SkyReelsModel="pretrained_models/SkyReels-V1-Hunyuan-I2V" python3 video_generate.py \ --model_id ${SkyReelsModel} \ --task_type i2v \ --guidance_scale 6.0 \ --height 960 \ --width 544 \ --num_frames 97 \ --prompt "FPS-24, A girl is dancing!" \ --embedded_guidance_scale 1.0 \ --image gsl.jpg

output:

FPS-24.A.girl.is.dancing._42_0.mp4

[pftq]

The steps (default 30 if not set) is not high enough for the guidance_scale (CFG) at 6. The sweet spots for me have been either:

• guidance_scale 3 and steps 20 or
• guidance_scale between 6-8 and steps 100.

If the motion is fast, you'll want to bump up the steps a bit more.

I posted some examples in response to another thread here:
kijai/ComfyUI-HunyuanVideoWrapper#367 (comment)

[pftq]

Some quicks tests with your image on better steps-to-guidance ratio to demonstrate what I mean:

40 steps, 3 guidance
(there's some fuzz/blur because it has lot of motion, so probably needed a few more steps than this)
https://github.com/user-attachments/assets/d3fb1958-bef9-40ac-9c99-4cd6837ab36c

100 steps, 6 guidance
https://github.com/user-attachments/assets/6b96e6a2-3e2c-497f-8b94-6f0127c98723

and doubling resolution makes a huge difference - makes the details come out way more and the motion more natural.
https://github.com/user-attachments/assets/a494f305-fbe1-4b62-838a-695435a75df9
(just a short clip sadly as this took hours to render even on H100 GPU)