https://tube.sethpc.xyz/videos/watch/ff08d0ba-3da9-4aa8-9def-70b3b97b0c09 Session 32 fix #3b: latent chain + scene-boundary B-injection. Mechanism (Seth's proposal): at each A→B scene boundary, generate B's clean latent via fresh txt2img and blend it into the chain over 6 frames with a ramp-down schedule — α_t = α_max · (1 - t/(N-1)), so first frame of window gets α=0.5 (max detail infusion when prompt changes), last frame gets α=0. Injects B's intact detail budget into the chain exactly when it's needed most. First scene has no preceding A — no B-injection there. 3 windows fired (scenes 1, 2, 3). Only variable changed vs iter 23 (pure latent chain): B-injection enabled. Everything else identical — SDXL Turbo, 6/12 inference steps, CFG 0.0, base_strength=0.25, 1024², 6fps→RIFE 24fps, EulerAncestral, no Gemma, same 4-scene Opus storyboard. A/B references on same storyboard: iter 20 base=0.25 png chain https://tube.sethpc.xyz/w/1ZcNNsdZQmyd4kjqaBVU38 iter 21 base=0.35 png https://tube.sethpc.xyz/w/kktv3muvTXnwz5BHYh5JBV iter 22 base=0.20 png https://tube.sethpc.xyz/w/fFgtTSsgNojMN3mFP8G83w iter 23 latent, no injection https://tube.sethpc.xyz/w/eHgTgQMia5yopNkJbm452U iter 24 latent + B-inj (this) (this upload) Wed, 17 Jun 2026 19:55:43 GMT https://validator.w3.org/feed/docs/rss2.html PeerTube - https://tube.sethpc.xyz https://tube.sethpc.xyz/lazy-static/avatars/8e6b7b73-cbe8-40ac-9236-7b0afee31de7.png https://tube.sethpc.xyz/videos/watch/ff08d0ba-3da9-4aa8-9def-70b3b97b0c09 All rights reserved, unless otherwise specified in the terms specified at https://tube.sethpc.xyz/about and potential licenses granted by each content's rightholder.