gelasdev-ComfyUI-FLUX-BFL-API/CHANGELOG.md

Changelog

All notable changes to this project from v1.1.0 onward are documented in this file. Earlier history lives in git log.

[1.7.0] — 2026-10-17

Added

Node / Feature	Endpoint	Notes
Flux 3 Image Klein (BFL)	POST /v1/flux-3-image-klein	BFL release 2026-10-15: the open-weights 5.6B FLUX 3 Image. Same prompt (bounding boxes as a JSON list at its end), image_1–image_10 references, resolution and aspect_ratio as Flux 3 Image, and no grounding. Adds seed (-1 = random), steps (1–16), output_format (jpeg / png), an exact width × height (64–6256, sent only when both are set, overriding resolution and aspect ratio) and safety_tolerance 0–7. Optional fields are sent only when changed. Shares Flux 3 Image's 120-attempt polling ceiling.
FLUX 3 Image Klein example workflow groups	—	workflows/BFL-API-flux-3-image.json gets two groups: Klein — Text to Image (bounding boxes, fixed seed) and Klein — Edit an Image (reference by URL). Appended only, so the existing nodes, links and groups are unchanged. The new groups are bypassed like the rest. Loaded in ComfyUI 0.37.0 (frontend 1.52.7).

[1.6.0] — 2026-10-02

Added

Node / Feature	Endpoint	Notes
Flux 3 Image (BFL)	POST /v1/flux-3-image	BFL release 2026-10-01. One node for text-to-image and editing: prompt (bounding boxes go at its end as a JSON list), up to 10 reference sockets image_1–image_10 sent as images and numbered Image 1…N in socket order (a skipped socket logs a warning, since the numbering shifts), resolution (768sq / 1k / 1.5k / 2k / 4k), aspect_ratio (15 ratios or auto), grounding (web and image search, on by default), safety_tolerance (0–4). The endpoint rejects seed, output_format and webhooks, so none are exposed. Polls up to 120 × 5 s because 4k can take several minutes, and keeps the result lossless (png) on its way into the IMAGE tensor.
FLUX 3 Image example workflow	—	New workflows/BFL-API-flux-3-image.json, built and saved with ComfyUI 0.37.0 (frontend 1.52.7): text-to-image with bounding boxes in the prompt, an edit with a reference given by URL (core Text node), and a style transfer that combines a local image (Load Image → Image to Base64, as Image 1) with a URL reference (Image 2). Groups are bypassed like the other examples.

Fixed

Issue	Detail
Polling ignored the returned polling_url	BFL requires polling the polling_url from the submit response (it points at the region holding the task; FLUX 3 Image returns a regional one). BaseFlux.post_request now passes it to get_result along with the task id, for every node; the self-built get_result?id= is only a fallback when a response has none.
Failed tasks reported as HTTP 503 were polled until timeout	FLUX 3 can report a failed task as HTTP 503 with a normal JSON body. A non-200 poll whose body carries Error, Content Moderated or Request Moderated now stops at once and logs the details instead of retrying to max_attempts.

[1.5.0] — 2026-09-26

Added

Node / Feature	Endpoint	Notes
Flux Video Edit (BFL)	POST /v1/flux-tools/video-edit-v1	FLUX Video Edit [fast]: video (MP4 URL or base64, ≤ 15 s / 50 MiB) + edit prompt + safety_tolerance (0–4). The endpoint rejects every other field (no seed, webhook or generation controls), so none are exposed. Duration, aspect ratio and audio follow the source; output is 720p. Outputs VIDEO, shares the Flux 3 Video polling ceiling.
Flux Video Upscale (BFL)	POST /v1/flux-tools/video-upscale-v1	BFL release 2026-08-20. input_video (≤ 20 s, 50 MB, 2560×1440), upscale_factor (1.5–3, default 2), creativity (0 precise / 1 creative, default 1), optional prompt, safety_tolerance (0–4), webhook. Outputs VIDEO.
Flux Deblur (BFL)	POST /v1/flux-tools/deblur-v1	BFL release 2026-06-25. Single image (base64 or URL, ≤ 4 MP), no prompt or mask; safety_tolerance (0–5), output_format (png default), seed, webhook.
qhd / uhd resolution on Flux 3 Video T2V / I2V / V2V	POST /v1/flux-3-video	BFL release 2026-09-10: 2K (2560×1440) and 4K (3840×2176) at 16:9 from a single request.
disable_pup on Flux 2 Max / Pro / Pro Preview	POST /v1/flux-2-{max,pro,pro-preview}	Turns off the automatic prompt upsampling these models apply by default. Only sent when enabled.
safety_tolerance and disable_pup on Flux Outpaint (BFL)	POST /v1/flux-tools/outpainting-v1	Both were in the API schema but not exposed. Only sent when changed from the defaults (2 / false).
Example workflow groups	—	BFL-API-tools.json: a Flux Deblur group (Load Image → Image to Base64 → Flux Deblur → Preview). BFL-API-flux-3-video.json: FLUX Video Edit and FLUX Video Upscale groups (Load Video → Video to Base64 → node → Save Video). Appended only — existing nodes, links and groups are unchanged; new groups are bypassed like the rest.
Moderation / error details in the polling log	GET /v1/get_result	On a terminal status (Error, Request Moderated, Content Moderated) the details payload is printed — for moderation it carries the Moderation Reasons categories added by BFL on 2026-06-15.

Fixed

Issue	Detail
Fresh Flux 2 nodes failed validation	All seven Flux 2 nodes defaulted width / height to 0 (meaning "not sent — BFL picks the size") but declared min: 64, so ComfyUI rejected every new Flux 2 node, and the Pro Preview / Klein 9B Preview example groups, with "Value 0 smaller than min of 64" until both were changed by hand. min is now 0 with a tooltip; 0 is still omitted from the request and any other value is sent as before.
Finetune example workflow needed a third-party node pack	BFL-API-finetune.json used ShowText|pysssss (ComfyUI-Custom-Scripts) to display results, plus an unused My Finetunes group-node template — both surfaced as "Missing Node Packs" on load. The four display nodes are now ComfyUI's core Preview as Text (PreviewAny), the stale template is removed, and the four utility groups are resized to fit. Verified in ComfyUI 0.37.0 (frontend 1.52.7).
Flux 3 Video V2V offered 16–20 s durations	BFL capped video continuation at 15 s on 2026-08-17; longer values returned 422 and a blank video. The V2V duration list now stops at 15. T2V / I2V stay at 20.
draft: true with a non-hd resolution was rejected	Drafts always render at hd and BFL rejects any other resolution with draft. The node now logs a warning and sends the draft at hd instead of wasting the request.

[1.4.0] — 2026-08-06

Added

Node / Feature	Endpoint	Notes
Flux 3 Video T2V (BFL)	POST /v1/flux-3-video	Text-to-video (mode: t2v), up to 20 s with synchronized audio. Knobs: resolution (hd / fhd), duration (auto or 5–20 s), aspect_ratio, generate_audio, safety_tolerance (0–4), draft (fast preview). Outputs ComfyUI's native VIDEO type; polling ceiling raised to 240 attempts (~20 min) for video — a real v2v+fhd task was still generating at 11 min.
Flux 3 Video I2V (BFL)	POST /v1/flux-3-video	Image-to-video (mode: i2v). Single keyframes string input: a bare image (base64/URL), a JSON array of images, or [seconds, image] pairs (up to 10). Warns when 3+ plain keyframes are sent with duration: auto (BFL requires a set duration). Same shared knobs as T2V.
Flux 3 Keyframes (BFL)	—	Utility: combines up to 10 image sockets (start_image, image_2–image_9, end_image) into the keyframes JSON string for I2V. Empty sockets are skipped. timing: even sends a plain list (first starts, last ends, middles spread evenly); timing: custom sends [seconds, image] pairs — the BFL schema allows no mixing, so start_image is auto-pinned at 0, middles use their time_N widgets, and end_image lands at end_time (with duration: auto that value is the clip length). Sorted into time order; warns on duplicate times or when end_time is not the largest.
Flux 3 Video V2V (BFL)	POST /v1/flux-3-video	Video continuation (mode: v2v) from start_video (MP4 URL or base64). Same shared knobs as T2V.
Video to Base64 (BFL)	—	Utility: converts a ComfyUI VIDEO input to a base64 MP4 string for V2V's start_video. Reads the stream source directly when it is already MP4 (in memory or on disk); other containers are remuxed to MP4 via VideoInput.save_to.
Flux Virtual Try-On v2 (BFL)	POST /v1/flux-tools/vto-v2	VTO v2 (BFL release 2026-07-17): sharper face preservation and garment detail, inputs up to 4 MP. Identical request/response format to v1 — implemented as a subclass overriding only the endpoint path.

Fixed

Issue	Detail
'Generating' is not a valid Status during polling	BFL's get_result reference documents two intermediate statuses the Status enum was missing: Reasoning and Generating (seen on FLUX 3 tasks). Both are now treated like Pending (wait 5 s, retry) instead of tripping the ValueError handler with a misleading "JSON parsing error" log.

[1.3.0] — 2026-06-25

Added

Node / Feature	Endpoint	Notes
mode on Flux Outpaint (BFL)	POST /v1/flux-tools/outpainting-v1	Quality/speed tradeoff added by BFL on 2026-06-09. Exposed as a combo defaulting to high (the API default); only sent when set to fast, so the default request body is unchanged.

[1.2.0] — 2026-06-01

Added

Node / Feature	Endpoint	Notes
Flux Virtual Try-On (BFL)	POST /v1/flux-tools/vto-v1	Dress a person image with a garment image. Base64 string inputs for person and garment, a required prompt, and optional safety_tolerance (0–5), output_format (jpeg / png), seed, webhook_url, webhook_secret. Follows the shared BaseFlux post → poll path. Two virtual try-on groups added to the tools example workflow.

[1.1.0] — 2026-05-25

Added

Node / Feature	Endpoint	Notes
Flux Erase (BFL)	POST /v1/flux-tools/erase-v1	Object removal via base64 image + binary mask. White pixels in the mask are erased; black pixels are kept. Knobs: dilate_pixels (0–25, default 10), safety_tolerance (0–5), output_format (png / jpeg), seed, webhook_url, webhook_secret.
Flux Outpaint (BFL)	POST /v1/flux-tools/outpainting-v1	Image extension to a target canvas. Knobs: width / height (≥64, step 32), center_reference (toggle), reference_offset_x / reference_offset_y, optional prompt, auto_crop, output_format (png / jpeg).
image_format on Image to Base64 (BFL)	—	New optional dropdown: jpeg (default, backward-compatible) or png (lossless, recommended for masks fed into Flux Erase / Flux Pro Fill).

Fixed

Issue	Detail
Workflow crash on non-multiple-of-32 width/height	BaseFlux.generate_image now catches check_multiple_of_32's ValueError inside its try/except and returns a blank image, matching the rest of the graceful-failure path.
FluxErase image / mask defaulted to None	Defaults are now "" — prevents JSON-null serialization when the socket is left unwired. Consistent with FluxOutpaint's input_image default.
Whitespace-only prompts on Flux Outpaint	Prompts that contain only whitespace are no longer forwarded — if prompt and prompt.strip(): strips the no-op case.