ComfyUI 0.45.4 adds Seedance 2.5 Flash and HeyGen Video 1.0
The release also puts three Meta Muse models and Claude Fable 5.2 on the OpenRouter and Claude nodes, and moves Krea and Wan to the Router backend.

Key takeaways
- ComfyUI v0.45.4 shipped on November 15, 2026 with partner nodes for ByteDance's Seedance 2.5 Flash (480p and 720p) and HeyGen Video 1.0 (up to 1080p and 30-second clips).
- Seedance 2.5 Flash bills 2.112 credits per 1,000 tokens with no reference video and 1.297 with one, and renders only at 480p and 720p.
- HeyGen Video 1.0 runs through two partner nodes, Reference to Video and Image to Video, and its auto duration option sets the clip length from the quoted dialogue.
- The same release adds three Meta Muse models (Wonder, Spark 1.3, Ember) and Claude Fable 5.2, retires Luma's legacy nodes, and moves Krea and Wan to the Router backend.
ComfyUI v0.45.4 turns two more video models into nodes you can drop into a graph. The release, dated November 15, 2026 in ComfyUI's changelog, adds partner nodes for ByteDance's Seedance 2.5 Flash and HeyGen's Video 1.0, and puts three Meta Muse models and Anthropic's Claude Fable 5.2 on the OpenRouter and Claude nodes.
Partner nodes are cloud API calls dressed as nodes. You sign in with a Comfy account, buy prepaid credits, and a node runs a hosted model with no local checkpoint and no API key to manage, according to ComfyUI's partner-node overview. The two new video nodes are the reason to update this week. Seedance 2.5 Flash is the cheaper, faster Seedance tier, and HeyGen Video 1.0 builds the picture, the spoken line and the ambience in one pass.
What ComfyUI 0.45.4 adds
The release note lists four partner-node additions and three changes to existing ones. The additions are Seedance 2.5 Flash at 480p and 720p, HeyGen Video 1.0 with 1080p output, 30-second clips and auto duration, plus three Meta models named Muse Wonder, Muse Spark 1.3 and Muse Ember, added to the OpenRouter LLM node, and Claude Fable 5.2 on the Anthropic Claude node.
The changes: Krea and Wan nodes moved to a Router backend, the Luma legacy nodes were removed (Ray 3.2 and UNI-1 remain), and Qwen3.6 Max Preview was retired from the OpenRouter LLM node.
Update ComfyUI before you look for the nodes
New nodes and their templates only appear after the core code and the frontend dependencies both update, so start with a full update rather than a code pull, per ComfyUI's update guide.
- On the portable Windows build, run
update_comfyui.batfrom theupdatefolder, which pulls the code and refreshes dependencies together. On a manual install,git pulland thenpip install -r requirements.txt. Restart ComfyUI withpython main.pyafterwards. - On Comfy Desktop, open Manage, then Update, and switch the channel to Latest on GitHub. Desktop tracks the stable release by default, and stable lags the commits that carry 0.45.4.
- When ComfyUI comes back up, open the Template Library and search the node name. If a node is missing, the docs point to an out-of-date build or a node that failed to import at startup.
Then sign in: partner nodes need a Comfy account with credits, bought under Settings, then Credits. The system is prepaid, so a run cannot overspend.
How to run Seedance 2.5 Flash
Search the Template Library for "Seedance 2.5 Flash: Reference to Video", the template ComfyUI's Seedance 2.5 page lists for the Flash tier. The node takes a text prompt and optional reference images, video or audio, and returns a clip with synchronized sound.
Flash is the cut-down tier and the docs are blunt about its ceiling: it generates 480p and 720p only, up to 30 seconds, and bills the same rate at both, per the partner-node pricing page. The full Seedance 2.5 node reaches 1080p and 4K; Flash does not. If you need a 1080p or 4K master, use the base node or the draft-mode workflow, which renders a cheap 480p preview first and a high-resolution final from it.
Flash is billed by tokens. The rate is 2.112 credits per 1,000 tokens with no reference video connected and 1.297 with one, and a 5-second 720p Flash clip with no video input comes to about 230 credits, by the pricing page's own worked example. A connected reference video cuts the per-token rate but adds its own duration to the token count, so a short 2-second reference can cost less than none.
Reach for Flash when you are iterating: a storyboard pass, a batch of variants, or a look test before you commit to a high-resolution render.
How to run HeyGen Video 1.0
HeyGen Video 1.0 is a general-purpose model that writes the picture, the spoken line and the room tone in one go, so a talking-head or product clip needs no shoot and no separate voice-over step. HeyGen's own ComfyUI guide splits it across two partner nodes.
- HeyGen Video 1.0 Reference to Video does text to video on its own, and reference to video once you connect images, clips or audio.
- HeyGen Video 1.0 Image to Video starts from a single first frame and keeps that image's aspect ratio.
Write the spoken line into the prompt in quotes, as part of the description of the shot, so the model speaks it and matches the lip movement. Describe the sound the same way: room tone, a music bed, or the noise an action makes is written into the prompt rather than set with a control. The prompt accepts up to 16,000 characters on both nodes, and reference media is addressed by name, as @Image1, @Video1 or @Audio1.
Reference to Video keeps people, products and places consistent across a shot. It accepts up to 9 images, 3 videos and 3 audio clips, 15 in total. The first two images are free and each one beyond that adds 6.03 credits. With aspect_ratio set to auto, the first reference image decides the framing of the video.
The docs now agree. The pricing page and the tutorial both describe clips of 5 to 30 seconds at 480p, 768p, 1080p or 2K. Two limits sit on top of that: a clip longer than 15 seconds needs no reference image and no reference video connected, and 1080p or 2K output needs a 16:9 or 9:16 aspect ratio. The pricing page adds an auto_duration option that follows the quoted dialogue instead of a fixed length, from 8 to 30 seconds, or 8 to 15 seconds with reference images or videos connected.
What HeyGen Video 1.0 costs
Prices are per second of output. On the current promotional rate, which ComfyUI says runs through December 1, 2026, Image to Video costs 3.02 credits per second at 480p, 5.43 at 768p, 12.07 at 1080p and 15.09 at 2K. Reference to Video is the same with nothing connected, and rises to 27.16 credits per second at 1080p or 2K once reference media is attached. A clip longer than 15 seconds moves to a higher long-clip rate on Image to Video. From December 1, the standard rates apply and the figures roughly double, so the same 1080p clip costs about 24.14 credits per second.
Meta Muse models and Claude Fable 5.2
The three Meta models land on the existing OpenRouter LLM node, so they appear in that node's model list rather than as new nodes of their own: Muse Wonder, Muse Spark 1.3 and Muse Ember. They are text models, useful in a graph for prompt expansion, shot lists and captions ahead of a video run, and they are billed at OpenRouter's token rates.
Claude Fable 5.2 is the new model on the Anthropic Claude node. ComfyUI's pricing page lists it at 2.414 credits per 1,000 input tokens and 12.069 credits per 1,000 output tokens, with cached input cheaper. That sits below the older Fable 5.1 and Fable 5 rates (3.017 and 15.087 credits) and above Opus 5.5 (1.207 and 6.035).
The rest of the release
Two backend moves matter if you have saved graphs. Krea and Wan nodes now run on ComfyUI's Router backend, a separate routing service from the partner nodes, which the pricing page prices separately. The Luma legacy nodes are gone, so any old workflow that used them breaks; Luma's Ray 3.2 and UNI-1 nodes remain. Qwen3.6 Max Preview disappeared from the OpenRouter LLM node's list.
The bug fixes are small and worth knowing: the Fish Audio Drama 3 node now shows the full price in its badge, and the Topaz video nodes and Kling Avatar 2.0 now show a request or task ID when a poll times out, so a stalled run can be told apart from a slow one.
None of this touches the nodes in SLOP TV's last ComfyUI how-to, which covered the 0.44.0 dubbing, lip-sync and camera nodes (sloptvnews.com). They are unchanged in 0.45.4.
Update to 0.45.4 (or the nightly), sign in to your Comfy account, buy credits under Settings, then Credits, and search the Template Library for "Seedance 2.5 Flash" or "HeyGen Video 1.0".
Sources
- docs.comfy.org - the v0.45.4 entry: node names, resolutions, models, date
- docs.comfy.org - Seedance, HeyGen, Claude Fable 5.2 credit rates and limits
- docs.comfy.org - what partner nodes are, account and credits requirement
- docs.comfy.org - Seedance node workflows and the Flash template
- docs.comfy.org - HeyGen Video 1.0 nodes, controls and prompting
- docs.comfy.org - how to update ComfyUI before using new nodes