nodes/alisson-anjos-ComfyUI-BFSNodes
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
237 Commits 14 Branches 0 Tags
dca37ddc0f943c7dded233da71c4bf7eedfb1cbc
Commit Graph
237 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
alisson-anjos dca37ddc0f Apply Colored Mark color to mask previews 2026-10-13 17:06:14 +00:00
alisson-anjos 507d5bc60a Always show Colored Mark region color 2026-10-13 16:59:56 +00:00
alisson-anjos 383fed81a1 Refine Colored Mark on updated main 2026-10-13 16:34:09 +00:00
alisson-anjos 2b0caac30a Fix Color Guide reference history and add global colors 2026-10-13 16:28:18 +00:00
alisson-anjos 27090c979a Add Shot Planner language selector and optional tour 2026-10-13 16:27:56 +00:00
alisson-anjos 3e2d112b03 Add optional Color Guide to Shot Planner 2026-10-13 16:27:29 +00:00
alisson-anjos 5f6e3c8446 Add five-frame shot history continuity 2026-10-13 12:38:58 +00:00
alisson-anjos 72af0aa0c7 Move shot filters to People Filters Masks tab 2026-10-13 07:33:51 +00:00
alisson-anjos d1ae8bee0a Bump BFSNodes version to 1.94.4 2026-10-13 07:26:11 +00:00
alisson-anjos 391115b78a Remove duplicate total duration control 2026-10-13 07:23:12 +00:00
alisson-anjos a55054aa0c Move max shots control to Video tab 2026-10-13 07:20:45 +00:00
alisson-anjos 154698f46f Re-split shots when cut settings change 2026-10-13 07:19:54 +00:00
alisson-anjos f1f639f389 Fix Shot Planner splitting and TextGenerate compatibility 2026-10-13 07:12:26 +00:00
alisson-anjos ceeff4803d Add simplified H3 shot conditioning 2026-10-12 13:13:43 +00:00
alisson-anjos 8aeb958ab6 Planner: stabilize VLM descriptions and default videos to 15s 2026-10-12 10:27:36 +00:00
alisson-anjosandClaude Opus 5.5 29cedad7be Shot loop: one-shot split mode, BFS Shot Seed, BFS Shot Noise, BFS Sigmas Preview (1.94.0) 
- Planner: Mode "One shot (whole range)" (the range as a single shot, Max seconds / shot not applied) and
  "Range length, seconds" next to it, so a whole shot or pieces of at most X are set in the Video tab alone.
- BFS Shot Seed: one seed per shot from a base seed (same / increment / random per shot) with per-shot overrides.
- BFS Shot Noise (experimental, optional): starting noise for SamplerCustomAdvanced on H3's video + audio latent;
  every amount at 0 is exactly RandomNoise. shared_base (the coarse pattern from a base seed, the same across shots,
  resolutions and lengths), static_share, distribution, spectrum_tilt, pyramid, guide_structure, ref_tint.
- BFS Sigmas Preview: chart of a sigma schedule (steps on layout / shapes / detail), pass-through.

Tested: split logic, seed logic, noise statistics (unit variance, identity at 0) and the chart, in Python.
Not tested yet inside ComfyUI / the panel in a browser.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-11 23:47:53 +00:00
alisson-anjosandClaude Opus 5.5 187c9a1776 Find people / Cast without insightface: OpenCV Zoo YuNet + SFace (Apache-2.0, run by OpenCV, downloaded once into models/opencv_face); InsightFace only when installed; SAM 3.1 + colours as last resort 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-09 19:00:36 +00:00
alisson-anjosandClaude Opus 5.5 a04f3116cf {details} filled automatically by the render nodes too: the planner leaves it when it has no VLM, BFS Shot H3 Conditioning / Shot H3 Duet (new vlm input) describe the shot's references 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-09 12:48:30 +00:00
alisson-anjosandClaude Opus 5.5 84a953e8c5 source_blur (SAM 3.1, keeps the mouth): blur the source person in the guide / native ref video / duet panel so the model reads pose, motion and lips but has no face to copy - Shot Conditioning, Shot Duet, Duet, Duet Conditioning 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-09 12:10:27 +00:00
alisson-anjosandClaude Opus 5.5 0bf03a8fb7 Planner: chunked upload for big files (8 MB pieces to /bfs/upload/chunk, assembled on disk) - any size, past ComfyUI's 100 MB --max-upload-size 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-09 09:03:16 +00:00
alisson-anjosandClaude Opus 5.5 390f9e8714 Planner UI: ✕ clear all - removes every reference from the plan (global, shots, cast) and the recent list, after a confirm 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-09 07:50:53 +00:00
alisson-anjosandClaude Opus 5.5 515ab73c70 Planner UI: collapsible sections keep what the user clicked (re-renders snapped them shut); ✕ clear for the recent reference pictures 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-09 07:47:46 +00:00
alisson-anjosandClaude Opus 5.5 2e64d989f6 BFS Shot H3 Duet: continuity 'history' too (the previous end before frame 0, panel strip filled with the source under it); shared shot_history() pops it from the shot 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-09 02:17:41 +00:00
alisson-anjosandClaude Opus 5.5 b1e0fcbbb6 Shot Planner: continuity 'history' is the default for new plans and unset shots (cuts still skip it) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-09 02:14:20 +00:00
alisson-anjosandClaude Opus 5.5 0b5315c45a History / chain: load the previous result on the CPU, slice uint8 before converting, keep history as uint8 in the shot and drop it after conditioning (no idle RAM/VRAM) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-09 01:39:46 +00:00
alisson-anjosandClaude Opus 5.5 a03e0be342 Shot Planner: continuity 'history' - the last 17/34/51 frames of the previous result placed just before frame 0 on H3's timeline (sliding history, after Akatz's H3 Relay / Ethanfel's Context Loop), with every mask mode, crop and the duet canvas; cuts reset it 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-09 01:37:36 +00:00
alisson-anjosandClaude Opus 5.5 78aa214584 Shot Planner: {details} described automatically before any SAM 3 work (it ran after the masks, which breaks the VLM: the button worked, the run did not); never leaves a hole in the sentence 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 23:03:47 +00:00
alisson-anjosandClaude Opus 5.5 05184747b2 Duet VLM prompt: place and look as lower-case noun phrases inside the sentences 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 21:32:31 +00:00
alisson-anjosandClaude Opus 5.5 596a1c2538 Duet prompts: TSC's newer strategy - the finished split screen as plain footage (no kept footage / generated / in sync), the pinned region named by place + one clause of what is seen ({panel_region}/{video_region}: top quarter, bottom three quarters...), motion beat by beat with times on the video clock, sounds of the new scene 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 21:18:07 +00:00
alisson-anjosandClaude Opus 5.5 4e4fc20fdd Duet: optional VLM prompt - writes it from the task, or fills the {...} fields of a hand-written template (TSC's); setting picture described as <Subject 2> 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 21:08:35 +00:00
alisson-anjosandClaude Opus 5.5 458099a2aa Duet: size refs for the whole canvas (as TSC's Reference to Video at canvas size) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 21:01:00 +00:00
alisson-anjosandClaude Opus 5.5 ad99a9f91a Duet: panel_static (TV static over face/person, SAM 3) + setting picture input + BFS Setting Picture node 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 20:52:18 +00:00
alisson-anjosandClaude Opus 5.5 f763b51779 BFS H3 Duet: side_by_side output - the same comparison as BFS Shot Join (original | references | result, info on top, prompt below) (1.83.0) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 20:21:36 +00:00
alisson-anjosandClaude Opus 5.5 2b462b639a BFS H3 Duet: megapixels input removed again (the size is computed in the workflow instead) (1.82.1) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 19:58:56 +00:00
alisson-anjosandClaude Opus 5.5 864a28b161 BFS H3 Duet / Duet Conditioning: megapixels - output size from the clip's aspect ratio at that area (0 = width / height) (1.82.0) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 19:54:02 +00:00
alisson-anjosandClaude Opus 5.5 e43cc14d9a Points: Track -> whole video - one selection tracked by SAM 3 through the whole video or range (chunked, forward and backward, re-found by text after cuts); shots without their own selection take their part (1.81.0) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 19:47:21 +00:00
alisson-anjosandClaude Opus 5.5 4fab22caa9 Shot Planner range: a player in the Video tab, start / end in frames (start here / end here take the player's exact frame; seconds of older plans still read), aligned layout (1.80.0) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 19:09:13 +00:00
alisson-anjosandClaude Opus 5.5 96f7e8a1b1 VLM descriptions: temperature and seed (0 = greedy as before), a dice button to try another description, the short preset capped at 30 words (1.79.1) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 18:02:53 +00:00
alisson-anjosandClaude Opus 5.5 34ffcef172 Shot Planner: Start at / End at range - only that part of the video is split, analysed, generated and joined; timeline dims the rest; max total seconds counts from the start (1.79.0) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 17:42:42 +00:00
alisson-anjosandClaude Opus 5.5 9ae5fc9614 Fit the prompt: only for the short preset, the VLM is told to reply with the blank's words only, and an echo of the prompt sentence is cut; Checks warn when {details} is used with a paragraph preset (1.78.2) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 15:14:58 +00:00
alisson-anjosandClaude Opus 5.5 7d5f60ce86 insightface and onnxruntime are optional: no pin in the requirements (the exact insightface pin and its own requirements conflicted with many environments); Find people / LTX Identity say what to install (1.78.1) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 13:50:54 +00:00
alisson-anjosandClaude Opus 5.5 8e573275b8 Fit the prompt: descriptions ({details}, {target}) are written for the prompt sentence that holds them (blank in place, article matched); {target} is described automatically at run time for shots with a selection and no text (1.78.0) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 13:34:08 +00:00
alisson-anjosandClaude Opus 5.5 a5535dc77c Describe refs for any subject (person, creature, robot, animal, cartoon: says what it is first); new default preset 'short' - a noun phrase that fits '<Subject 1> is the {details} shown in <Picture 1>'; {target} describes animals/creatures too (1.77.0) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-08 11:35:11 +00:00
alisson-anjosandClaude Opus 5.5 7af3fa5618 pose look: clearer error (package vs model, with the download link) and no restart needed after adding the model (1.76.1) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-07 16:07:09 +00:00
alisson-anjosandClaude Opus 5.5 39610ab1ef mask_guide_look: what the mask guide shows of the subject - grey blurred (default), colour, silhouette, edges, or pose (people, YOLO pose model from models/ultralytics, no new packages) (1.76.0) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-07 15:47:14 +00:00
alisson-anjosandClaude Opus 5.5 aa92d72990 Faster masks: separable dilation / blur (same result, ~7x faster on few CPU threads) and the generation mask computed once per shot (1.75.1) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-07 15:35:14 +00:00
alisson-anjosandClaude Opus 5.5 f1388f0463 mask_guide (renamed from person_guide, works for any subject): masked region as an extra guide, instead of the full guide, or as a native reference video (<Video n>, with mask_ref_size); Mask only shots only; Frame + paste takes a per-shot 'new subject' text for the new outline (1.75.0) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-07 14:51:56 +00:00
alisson-anjosandClaude Opus 5.5 5c93d23c88 H3 Conditioning: experimental person_guide - a guide latent with only the masked person (rest grey), added next to the full guide or instead of it (1.74.0) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-07 14:45:55 +00:00
alisson-anjosandClaude Opus 5.5 7d9d7a9e26 Frame + paste mode: the whole frame is generated (pose follows the guide, the new person may be bigger) and BFS Shot Join pastes only the person (shot mask + the new outline found by SAM 3) onto the original frame; mask strength renamed mask opacity (1.73.0) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-07 14:22:16 +00:00
alisson-anjosandClaude Opus 5.5 1de7753be9 Mask strength per shot (mask modes): the masked rows start at strength x the noise level and keep part of the original, like an opacity (1.72.0) 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-07 14:09:17 +00:00