Repository navigation
Releases
v0.11.0
New
- Avatar Generation native template: Seedream 5.5 creates a portrait, Seed Audio creates narration, and OmniHuman produces the avatar video in one connected run.
- Includes portrait, audio and video savers, a completed task-ID preview, and optional portrait/reference/audio upload nodes.
- Workflow guide, native graph thumbnail and downloadable template included.
Fixes And Guidance
- OmniHuman cancellation recognizes HTTP 400 with integer code 50217 as an "already started" refusal, without reporting successful cancellation.
- Audio tooltips and docs explain that video duration follows the input audio. Trim or regenerate narration to change video length; there is no separate duration parameter.
Requirements
- ComfyUI 0.31.0 or newer. ModelArk, Seed Speech and Vision AI each require their own BytePlus key.
- Connected OmniHuman media uploads require Comfy.org sign-in.
- One Run executes the template without an approval pause. To review assets first, use selected output-node execution or the separate Avatar Studio extension.
- Optional upload nodes start muted and disconnected. When replacing a generated input, mute its generator and saver to avoid unused generation.
Validation
- 711 local tests passed with no skips; native HTTP registration/template checks, documentation build, lint and archive validation passed.
- Default, reference-image and upload-only paths were previously exercised with mocked providers and real native savers on installed and minimum ComfyUI.
- No additional paid generation was performed for this release.
Registry upload success is separate from security-scan approval. Check the version status before relying on Manager availability.
v0.10.0: OmniHuman 2.0 and five-minute audio
What's New
- OmniHuman 2.0 image-plus-audio avatar generation, Task Result collection, and explicit queued-task cancellation.
- Audio support from 4 to 300 seconds, inclusive. Oversized connected WAVs are downmixed to mono and resampled to at most 24 kHz without truncation; fitting clips retain their rate and channels.
- Separate Vision AI credential in Settings > BytePlus (
BYTEPLUS_CV_API_KEY). ModelArk, Speech and MediaKit keys are not substitutes. - Private atomic task recovery, no automatic paid-submit retries, bounded transient polling retries, and public-only credential transport and redirect-safe video downloads.
Validation
709 local tests passed without skips. Main CI, docs deployment, and the tag/Registry-upload workflow passed. Every one of the 114 uploaded Registry package files byte-matches commit c0ffd24f62bbe11697c7b96a1a398f3e129a3d20.
Five-minute connected-audio preparation and a real ComfyUI workflow with mocked remote calls passed. The earlier paid live generation passed at four seconds; paid five-minute generation, live Comfy.org uploads and queued cancellation remain separate validation follow-ups. No additional paid generation was made for this release.
Updating
The Registry upload succeeded. At publication on November 19, 2026, version 0.10.0 is Pending security review; this is not confirmation of availability in ComfyUI-Manager. Check the Registry version status before updating through Manager. Do not bypass pending or flagged review. Restart ComfyUI and reload the browser after updating.
See the update guide and OmniHuman guide.
What's Changed
- Add OmniHuman 2.0 with five-minute audio support by @ArthurReimus-ByteDance in #35
Full Changelog: v0.9.2...v0.10.0
v0.9.2
What Changed
- Seedance 2.5 Flash exposes and forwards MP4/MOV in all three generation nodes. Invalid formats fail before submission.
- Flash references are limited to 10 images, 5 videos, and 5 audio clips (20 total), counting connected media and asset references together.
- Automatic 2.5-family omni-reference tasks use adaptive ratio and duration -1, avoiding conflicts with inferred editing. Explicit reference/edit/extend behavior and non-blocking task reuse are preserved.
- Seedream 5.5 guidance reflects its official release. Model IDs, resolutions, defaults, and existing security safeguards are unchanged.
- All 39 example workflows and documentation downloads now carry the v0.9.2 BytePlus node version.
Verification
- 664 local tests passed without skips; isolated real-ComfyUI queue smoke checked six real SDK video request serializations with mocked provider transport.
- Lint, formatting, Python compilation, JavaScript syntax, strict docs, 86-page docs build, package validation, diff review, and secret-pattern checks passed.
- Main CI and the official tag/publish workflow passed. All 111 files in the downloaded Registry package match the tested release commit.
- Two independent reviews plus follow-up reviews found no remaining actionable issues.
No new paid provider generation was performed. Flash seed determinism and live acceptance of the newly corrected request cases remain unverified. No standalone Python type checker is configured.
Installation Status
The package was uploaded to the Comfy Registry. Its initial verification status on 2026-11-16 was NodeVersionStatusPending: upload success does not establish security acceptance or availability in Manager. Check the live Registry version status before updating. Once active, update through ComfyUI Manager and restart ComfyUI.
Release commit: ac2089a.
v0.9.1 - Credential and Reference Security Hardening
Security Hardening
- ModelArk key validation uses public-only networking and fixed regional endpoints, disables redirects, and logs exception classes without credential text.
- Reject malformed API keys before networking or persistence. Restrict credential-bearing shared HTTP requests to BytePlus HTTPS provider origins.
- Validate reference slots and probe redirects, rejecting embedded credentials, private addresses, control characters and malformed destinations. Connected DNS checks protect against rebinding.
Verification
653 tests passed without skips on both local ComfyUI and current master. PR, main and publishing CI, real ComfyUI credential-route smoke, lint, compilation, JavaScript syntax, Registry validation, strict docs/build, diff and secrets checks passed. No extra paid generations were made. All 78 workflows changed only BytePlus version metadata.
Seedream 5.5, standard Seedance 2.5 4K, and Flash support remain intact. Premium and Premium Draft remain removed. Flash retains the approved API-default format and 30/10/10 references.
Registry 0.9.1 is initially Pending review; v0.9.0 remains Flagged. Upload is not security acceptance or Manager availability. Do not bypass review. The docs-only Astro advisory remains separate follow-up work; no standalone type checker is configured.
PR: #33
Upgrade guide: https://byteplus-sa.github.io/ComfyUI-BytePlus-ModelArk/get-started/update/
Registry status: https://api.comfy.org/nodes/ComfyUI-BytePlus-ModelArk/versions?include_status_reason=true
v0.9.0 - Seedance 2.5 4K and Flash
Highlights
- Standard Seedance 2.5 now offers 480p, 720p, 1080p and 4K through
resolution, usingdreamina-seedance-2-5-260628at every resolution. - Seedance 2.5 Flash (
dreamina-seedance-2-5-flash-261115) is available in the existing Text to Video, First-Last-Frame, Reference to Video and task-query nodes. It supports 480p/720p, 4-30 seconds and 30-image/10-video/10-audio reference limits. Private-model access and activation are required. - Flash uses the API's default output format and has no Draft option. Standard Seedance 2.5 keeps MP4/MOV output, 480p Draft and 1080p Draft-to-Final.
- Documentation, migration guidance and all 39 shipped/downloadable templates are synchronized to this release.
Upgrade Notes
Premium and Premium Draft are removed entirely, with no compatibility aliases or automatic migration. Replace a Premium selection with Seedance 2.5 and resolution=4k. Old Premium drafts cannot be converted: create a standard Seedance 2.5 Draft for a 1080p final, or generate 4K directly.
The standard Seedance 2.5 endpoint applies at every standard resolution, including 4K, and does not override Flash. Saved keys stay unchanged. Restart ComfyUI after updating.
Publication And Verification
Released commit: c63e295d5f424e9e9513d186cb3a38f5daf8f170. All 636 local tests passed without skips; lint, fresh ComfyUI schema smoke, the docs build, main CI and release tests passed. All 111 files in the uploaded Registry package byte-match the tag.
The Registry upload succeeded, but version 0.9.0 is pending security review at publication. Check the Registry's Versions tab for acceptance before expecting a Manager update; do not bypass pending or flagged review.
Upgrade guide | Registry | Publish workflow
What's Changed
- Add unified Seedance 2.5 4K and Flash support by @ArthurReimus-ByteDance in #31
Full Changelog: v0.8.0...v0.9.0
v0.8.0 - Seedream 5.5
Seedream 5.5
Select seedream 5.5 in the existing Seedream or Seedream Layer Separation node to use early-access dola-seedream-5-5-261028. Existing model defaults and node IDs are unchanged.
- Text-to-image and editing with up to ten reference images.
- Adaptive 1K, 1.5K and 2K, fixed presets and custom output dimensions.
- PNG/JPEG with auto, transparent or opaque backgrounds; PNG alpha is preserved in the mask output.
- Standard prompt optimization and optional hashed end-user safety identifiers.
- Layer separation with a 2-100 total-image ceiling, including the base (default 30), and optional rich-text OCR in
layers_json.
Fixes
- Omitted optional ComfyUI inputs now correctly use their defaults instead of failing validation.
- Public-only result downloads block private addresses, unsafe redirects and DNS rebinding across image, video, 3D and MediaKit paths.
- Shared service requests no longer follow redirects automatically with credentials. Download retries, interrupt cleanup and stall timeouts are preserved.
- Documentation dependencies updated with compatible fixes; npm audit reports zero vulnerabilities.
- Updated ComfyUI core discovery names retain historical aliases; shared Seedream 5.5 schemas are checked without changing existing model defaults or verified BytePlus capabilities.
Documentation
Updated node and upgrade guides explain transparent PNG saving through Join Image with Alpha, the ten-reference ImageBatch reload workaround, optional OCR and compositor limits.
Custom output area is 921,600-4,624,220 pixels, not 4K. Layer separation is generative rather than lossless and can return fewer images than the ceiling. The current ComfyUI Create Layered Image compositor accepts at most 50 layers including the base; save/process larger sets separately. Review generated text before publication.
Verification And Update
622 local tests passed without skips against older ComfyUI, latest release v0.45.0 and current master. Ruff, JavaScript syntax, Python compilation, strict documentation checks, the 86-page documentation build, Comfy Registry metadata/package validation and a no-cost SDK/download/layer/alpha smoke test passed. Previously approved paid ComfyUI image and layer tests succeeded in Singapore.
GitHub CI passed on Python 3.10/3.12, including release/master core parity; the publication workflow also passed its release-only test gate. Updated guides are deployed. No standalone type checker is configured. A fresh integrated-browser recheck was blocked by HTTP 403; previous native workflow and paid E2E evidence remain separate from the no-cost runtime smoke.
Requires ComfyUI 0.31.0 or later. Seedream 5.5 still requires account early-access entitlement; Europe availability was not live-tested. The seed widget is a local rerun control, not a Seedream 5.5 API seed.
Registry upload and security review are separate. At publication verification on 2026-11-12, v0.8.0 was Pending review, not yet confirmed available in Manager. All 111 files in the uploaded package matched tagged commit 12c65f8f6ba9c9bb04bf37ee888cceed0acc7d4e. Check the current Registry version status before expecting the update in Manager; do not bypass pending or flagged review.
Seedream guide | Layer separation guide | Update guide | PR #29
v0.7.0
What's Changed
- Add default ModelArk endpoints and prepare v0.7.0 by @ArthurReimus-ByteDance in #27
Full Changelog: v0.6.1...v0.7.0
v0.6.1
What's Changed
- Fix interrupt/billing safety, redact signed URLs, harden CI and publishing by @ArthurReimus-ByteDance in #22
- Refactor: shared request, download and task layers by @ArthurReimus-ByteDance in #23
- Release v0.6.1 by @ArthurReimus-ByteDance in #24
Full Changelog: v0.6.0...v0.6.1
v0.6.0
What's Changed
- Add Hyper3D Gen2 and Hitem3D 2.0 nodes, 3D templates and release v0.6.0 by @ArthurReimus-ByteDance in #21
Full Changelog: v0.5.5...v0.6.0
v0.5.5
What's Changed
- Add production asset templates and release v0.5.5 by @ArthurReimus-ByteDance in #19
Full Changelog: v0.5.4...v0.5.5