What it does This extension turns a short video into a stylized or edited output by running it through a rigid seven-stage pipeline inside the AUTOMATIC1111 web UI. It extracts frames, picks keyframes based on motion density, runs img2img on those keyframes, then hands off to EbSynth for temporal propagation. No After Effects required.

The interesting bit The extension tries to automate the tedious parts: it varies keyframe spacing by motion (tighter where things move fast), can auto-tag frames with DeepDanbooru to catch blinks and hand gestures, and optionally applies LoRAs or embeddings dynamically based on tag strength. There’s even a face-crop img2img mode that processes faces separately to preserve expression detail.

Key highlights

• Seven-stage wizard-style workflow (frame extraction → keyframe selection → img2img → upscale → EbSynth project generation → EbSynth run → final video)
• ControlNet integration recommended; multi-ControlNet (canny + normal map) is explicitly supported with per-stage weight overrides
• CLIPSeg masking for selective editing (e.g., “cat” but exclude “finger”)
• Auto-tagging with DeepDanbooru, plus dynamic LoRA/prompt injection based on tag confidence with a blacklist for false positives
• Face-crop img2img with YuNet detection and separate denoising strength for faces vs. full frame

Caveats

• Progress during generation is not reflected in the Web UI; you watch the console like it’s 2005
• The README warns that current Web UI versions throw an error unless you drop an image on the main img2img screen, even though it “does not affect the result”
• Stage 3 is described as “the most time-consuming part” and requires manual trial and error with prompts, denoising strength, and face crop settings
• The 30-minute benchmark for a 15-second 720p clip is from a specific hardware/config setup and may not generalize

Verdict Worth a look if you’re already deep in the AUTOMATIC1111 ecosystem and want to batch-process stylized shorts without touching Adobe’s subscription stack. Skip it if you want real-time preview, clean error handling, or a tool that works outside the Stable Diffusion Web UI bubble.