Anime LoRA Workflow & Videos
Budget: $250 – $750 USD
I need a repeatable ComfyUI-based workflow that lets me train LoRAs in a manga art style, with focus on NSFW material, and then use those LoRAs inside any text-to-image generator for reliable character creation. The primary subjects will be fantasy heroines and everyday slice-of-life personalities; I want them to stay recognizable across multiple prompts and scenes.
Once stills are coming out consistently, I also want a simple route to turn the image series into short videos—nothing feature-length, just social-media clips assembled smoothly from the generated frames. If you prefer AnimateDiff, Deforum, or an FFmpeg node chain inside ComfyUI, that’s fine as long as it stays inside the same toolset and I can reproduce the results locally.
Deliverables
• A .json CompfyUI graph (or set of graphs) covering:
– dataset preparation steps, tagging guidance for NSFW compliance, and recommended hyper-parameters for LoRA training
– loading and testing the finished LoRA inside a text-to-image pipeline, with example prompts that keep character identity intact
• A companion graph or script that strings the exported frames into a smooth video (mp4), including any FPS or interpolation settings you’ve found effective
• A short README outlining installation prerequisites, exact node versions, and a step-by-step usage guide
Acceptance criteria
• I can feed my own dataset, press run, and obtain a LoRA that matches the manga style while handling NSFW content without crashes or color shifts
• Generated characters remain visually consistent across at least five different prompts
• The video pipeline renders a 5–10 second clip from the stills in one pass inside ComfyUI
Keep the setup Windows only, and please comment any custom nodes you rely on. If something needs a paid license, flag it early so I can decide.
Once stills are coming out consistently, I also want a simple route to turn the image series into short videos—nothing feature-length, just social-media clips assembled smoothly from the generated frames. If you prefer AnimateDiff, Deforum, or an FFmpeg node chain inside ComfyUI, that’s fine as long as it stays inside the same toolset and I can reproduce the results locally.
Deliverables
• A .json CompfyUI graph (or set of graphs) covering:
– dataset preparation steps, tagging guidance for NSFW compliance, and recommended hyper-parameters for LoRA training
– loading and testing the finished LoRA inside a text-to-image pipeline, with example prompts that keep character identity intact
• A companion graph or script that strings the exported frames into a smooth video (mp4), including any FPS or interpolation settings you’ve found effective
• A short README outlining installation prerequisites, exact node versions, and a step-by-step usage guide
Acceptance criteria
• I can feed my own dataset, press run, and obtain a LoRA that matches the manga style while handling NSFW content without crashes or color shifts
• Generated characters remain visually consistent across at least five different prompts
• The video pipeline renders a 5–10 second clip from the stills in one pass inside ComfyUI
Keep the setup Windows only, and please comment any custom nodes you rely on. If something needs a paid license, flag it early so I can decide.