ComfyUI

What this means for you: ComfyUI now includes Pruna P-H3-Image, a model that makes images from text. It is a distilled model, meaning a slimmer version built from a larger one, built on Minimax H3. You can run it on your own computer or in the cloud.

How to use it: Available now in ComfyUI. Run it locally on your own computer or on cloud.

What this means for you: Kling 4.0 makes video clips of 3 to 30 seconds at 720p or 1080p from text, an image or reference material, with audio built into every clip. There is also a node for editing video. Image to Video takes up to 10 keyframes, which are key pictures the video passes through.

Why it matters: It comes a day after ComfyUI added Pika H3 Flex on Oct 30, and adds to a run of video tools that began with Flux 3 Max on Oct 21.

How to use it: Kling 4.0 is now in ComfyUI as four nodes: Text, Image & Reference to Video, plus Video Edit. Use the link in ComfyUI's post to download the workflows.

What this means for you: ComfyUI, a free tool for building image and video workflows, can now make a 15 second video clip with audio using Pika H3 Flex. You can give it up to 12 reference images or clips to guide the result. You can draft at 480p or 768p and finish at up to 4K.

How to use it: Pika H3 Flex is available in ComfyUI now. Click the link in ComfyUI's post to try it.

What this means for you: Takes a video of people talking and gives you a version in another language. Each speaker keeps their own cloned voice, and the lips are matched to the new words while tone and emotion stay. The Pro tier adds more precise lip-sync, keeps the original background audio and handles videos with several speakers.

Why it matters: It follows a run of ComfyUI additions in voice and video, including HeyGen Voice Turbo on Oct 21 and new partner nodes for HeyGen Voice on Oct 27.

How to use it: VEED Dubbing is now available in ComfyUI. Pro is the tier for multi-speaker videos, precise lip-sync and keeping the original background audio.

What this means for you: This release adds a control for color depth in the MOSS-image model. It also adds partner nodes, which are blocks that connect ComfyUI to another company's model. The new ones cover FLUX 3 Max, HeyGen Voice (voice cloning and text to speech) and the Crystal Video Upscaler.

Why it matters: It follows v0.42.0 on Oct 18, which added the MOSS-image model. The FLUX 3 Max and HeyGen Voice nodes arrive after both were announced on Oct 21.

How to use it: ComfyUI v0.43.0 is released on the Comfy-Org/ComfyUI releases page on GitHub. ComfyUI is free and open source.

What this means for you: ComfyUI has released a test add-on (a LoRA, a small file that tweaks an existing image model) for the MOSS Image model. It helps make sharper, higher resolution images, up to 10 megapixels (MP). It leans toward anime, though the training included many photographs.

Why it matters: It follows ComfyUI adding support for the MOSS-image model in v0.42.0 on Oct 18, so you can now pair that model with this add-on.

How to use it: Download the LoRA from Hugging Face at huggingface.co/comfyanonymous/MOSS_image_MAX_ULTRA_hires_lora_test and use it with MOSS Image in ComfyUI. The team asks for feedback on whether to make more of these.

What this means for you: Flux 3 Max makes short videos from inside ComfyUI. It generates a 10-second video at 720p or 768p in about 12 seconds. The result is a video clip.

How to use it: Open the link in ComfyUI's post to run the Flux 3 Max workflows. A workflow is a saved chain of steps that ComfyUI follows.

What this means for you: HeyGen's Voice Turbo lets you turn text into speech inside ComfyUI, and clone a voice from a sample. It produces spoken audio. You pick a Turbo or Standard version of the text to speech.

How to use it: There are three pieces. The Instant Clone node is free during HeyGen's preview. The Text to Speech node lets you pick Turbo or Standard. The HeyGen Voice 1 template clones a voice and speaks in one workflow. Standard is 50% off through Oct 31.

What this means for you: ComfyUI can now use FastH3 OmniRef, a model that makes videos from several reference inputs, such as images, and runs on your own computer. It is a distilled version of Minimax H3, meaning a smaller, faster copy trained to imitate the larger model. The weights are open, so they are free to download.

How to use it: The ComfyUI post links to the open weights and to a cloud option. Run it locally on your own hardware, or use the cloud link.

What this means for you: ComfyUI, the free node editor for building image, video and audio workflows, can now use the MOSS-image model. It also has a Krea 3 Image node you can place in a workflow. This version also includes a reworked tool for remeshing 3D models and a fix for the webcam node, which had stopped working.

Why it matters: It follows Oct 15, when ComfyUI first added Krea 3 through Partner Nodes for generating and editing images from multiple reference images.

How to use it: Update ComfyUI to v0.42.0. The Krea 3 Image node is listed under Partner Nodes.

What this means for you: Pika video generation can now be reached through Comfy Router, so one API key (a code that lets software use a service) and one bill cover it. It sits alongside the other models you already route through ComfyUI.

How to use it: Click the link in ComfyUI's post to get an API key for Comfy Router.

What this means for you: ComfyUI can now run Flux3-image models. It also gains an ImageSDRToHDR node (a building block that turns a standard image into an HDR one) and a ReferenceMultiLatent node.

Why it matters: It arrives the same day ComfyUI announced native support for FLUX 3 Image Klein, which brings open weights, 4K generation and multi-reference input.

How to use it: Version v0.41.0 is posted on ComfyUI's GitHub releases page.

What this means for you: ComfyUI now runs FLUX 3 Image Klein, an image model you can download and run yourself ("open weights" means the model files are public). It generates images natively at 4K, takes several reference images, edits images, and renders and edits text inside pictures. Bounding boxes let you mark where things go in the layout.

Why it matters: ComfyUI v0.41.0, also released Oct 15, adds Flux3-image model support.

How to use it: Both the base and distilled versions have open weights. Use the link in the comments of ComfyUI's announcement post to download the weights.

What this means for you: You can give Krea 3 several reference images and tell it which one to use for what, by upload order, such as the person from image 1 and the jacket from image 2. You can swap a fabric or finish while the angle and room stay the same. You can then feed the result back in and keep editing.

Why it matters: ComfyUI also added Nano Banana 2.1 on Oct 6, with up to 14 reference images and multi-turn edits. Krea 3 is another reference-driven option in the same Partner Nodes.

How to use it: Krea 3 comes through ComfyUI's Partner Nodes, which are nodes that use models run by other companies.

What this means for you: ComfyUI can now generate a whole multi-character audio scene in one go with Fish Audio Drama 3. You can edit the scene or the audio without starting over. It also works alongside Fish Audio's text-to-speech, voice cloning and transcription.

How to use it: Fish Audio Drama 3 runs as a Partner Node inside your workflow. Partner Nodes are nodes that connect ComfyUI to outside companies' models.

What this means for you: Three new building blocks (called nodes) let you finish video footage on your own computer. One sharpens and cleans up low-quality video (upscale, denoise, deblur), one adds new in-between frames, and one converts standard video (SDR) to HDR. You can chain all three, and EXR files (a high-quality image format used in visual effects) are supported.

How to use it: Needs an NVIDIA RTX graphics card, since the nodes run locally on your RTX GPU. Click the link in ComfyUI's post to run the workflows and read their blog.