Show & Tell: Complete 12-episode AI donghua series built with ComfyUI + SDXL

[chfr19820610-cell]

Using ComfyUI + SDXL to produce a complete 12-episode AI donghua (animated web series)

I built a full AI animation pipeline around ComfyUI and just finished a complete 12-episode Xianxia-style series. Sharing the workflow in case it's useful to others exploring AI animation.

What I used

ComfyUI + SDXL 1.0 for base frame generation:

• Custom LoRA trained on 30 character reference images for consistency
• 15+ controlled keyframes per episode
• Resolution: 1080x1920 (vertical format)
• Sampler: DPM++ 2M Karras, 30 steps

Pipeline overview

1. ComfyUI generates SDXL frames (batch of 10-20 per scene)
2. Blender NPR shading for the anime/donghua look
3. Ken Burns effect + parallax layers for animation
4. TTS voiceover with SSML control
5. Remotion for final assembly + subtitles + BGM

Results

12 complete episodes, each 1-5 minutes. The key challenge was character consistency — solved with a combination of LoRA training and controlled keyframe generation.

Free 30-second sample (no signup): https://chfr1982.gumroad.com/l/hrscll

The full production pipeline is open-sourced here: https://github.com/chfr19820610-cell/AIComics

Happy to share workflow details, LoRA training tips, or answer questions about the pipeline!