Seeking a local/free workflow for 5-shot character-consistent video (Krea + H3 + ComfyUI)
NewPhoneWhotiz · reddit · 2026-09-22
A student wants a local/free workflow for a 5-shot video with cross-shot character consistency and storyboarding before rendering. Image side: Krea realtime canvas. Video side: MiniMax H3 + ComfyUI. Proposed pipeline: Krea 3 / Qwen for character and scene refs → H3/LTX for video → ComfyUI/editor for assembly. Asks whether this works or if there's a better way without training a LoRA.
More from Multimodal
- Distortion + inpainting + regeneration loop makes trippy AI art — PurzBeats · 2026-09-23
- Creator Pushes 'Bending Reality' Limits in AI-Generated Short 'It Wants You To Stay' — AIandDesign · 2026-09-23
- Unsloth GGUFs run Qwen-Image-2.1 locally on as little as 6GB VRAM — danielhanchen · 2026-09-23
- Generating a dubstep track whose lyrics are only 'cat' and 'meow' with Suno — thatroblennon · 2026-09-23
- zen-image-edit: open image-edit model matches Qwen-Image-2.1 with 17GB less memory — recoilme · 2026-09-23
- Version ComfyUI workflow JSONs as local git repos so your agent can roll back — PurzBeats · 2026-09-23