RedNode Studio is my custom-built creative studio for ComfyUI, designed for painters, designers, and AI artists; from first-time users to experienced workflow builders.

For more advanced text: https://github.com/RedNodeAI/ComfyUI-RedNodeStudio

🚀 Start here. Read this bit, it saves you an hour.

There are two workflows and you want the boring one first. RedNode Start Here worknode in the example work flow folder is the one you want to load 1st. It renders nothing. All it does is hold one node from every optional pack, so ComfyUI Manager can see the whole list at once and install it in one go. Skip it and you get a graph full of red missing nodes.

1. Install the pack. ComfyUI Manager, search RedNode Studio, Install (or Update if you are on 1.3), then restart ComfyUI.

2. Open the template browser (Workflow, then Browse Templates) and load RedNode Start Here Install. It will look like a page of unrelated nodes. That is the point.

3. Let Manager do the rest. ComfyUI Manager, Install Missing Custom Nodes, install everything it lists, then restart ComfyUI again.

4. Now open the real one. Template browser again, load RedNode Studio (the V1.4 template). This is the workflow you actually use.

5. Get the model files. There is a note inside that workflow listing every model and LoRA it loads and where each one comes from. The full list is further down this page too.

Not sure what is still missing? Open the Workspace's Overview tab. Check Installs at the bottom names every pack and model your current settings need, ticks the ones you already have, and links the rest. It installs nothing itself, Manager does that, so you can see what is happening.

Update 1.4.3
1.4.3 is out, and it is a big one: everything since 1.3 is in it. ComfyUI Manager works again; so please update from there.

Search RedNode Studio in ComfyUI Manager, update, restart. No more installing from GitHub. Then load the new RedNode Studio template (V1.4): it replaces the V1.3 workflow.

The big changes

• One node does the whole job now. The Detailer, Post FX and Save run inside the Workspace, so there is nothing to wire after it. The V1.3 chain of nodes is gone from the template.

• Overview tab. The whole run as boxes in the order it happens: green on, grey off, amber stood aside, red wants fixing, with the reason on hover. Click a box to open its page, right-click it to switch that stage on or off. Workspace Presets live at the top.

• Run tab. Generate, every stage with its time, the live picture, a log, and your VRAM as a chart while it renders.

• VRAM Limit. Set it to your card and the run holds models to stay under it. Same prompt on my 32 GB card: 22.7 GB uncapped in 1:12, or 7.4 GB capped at 8 GB in 1:16. A rig that would not fit your card renders on it anyway.

• Nothing fails silently. Errors say what went wrong and where to fix it, and Check Installs on the Overview names every node pack and model your settings need, with a link to each.

• The panel is easier to read. The big tabs are pages of sub-tabs, each with a light that says whether it is on, over a status bar that sums the page up.

• Draft switch. Skips the Detailer and Post FX so a seed is cheap to judge.

Models and rigs

• A rig can be your own nodes. Three small Rig nodes let your own loaders and sampler stand in as a rig, with no wires to the Workspace.

• A rig per pass can be another model family. Draft on Z-Image and finish on Krea 2, or the other way round, straight from the Latent or Img2Img tab.

• Swap works on any render, not only on an Img2Img source, and Re-angle and Swap now share one shape.

Prompts

• Import Prompt. Drop any picture on the Prompts tab and it reads the prompt and settings back out. A picture RedNode Studio saved gives back its whole setup.

• As Queued Last Run. The exact words the last run sent, every wildcard and socket resolved, with the seed and a Copy button.

• The row you chose is the row that renders, Rewrite has a Fresh Variation choice, and a Text Size slider scales the words.

• Full screen now shows the generation data beside the picture: prompt, settings and LoRAs, with Copy.

Detailer

• Simple view. It was starting to look like a cockpit. Simple shows the few dials most passes need, Advanced shows everything, and switching changes nothing in your settings.

• Tiled Upscale pass through Ultimate SD Upscale, and the SeedVR2 pass can now aim at a region, so a face gains the detail and the frame keeps its size.

• Blend, Tone Lock and Free VRAM on every pass: how much of the repaint goes back, colour that stops drifting down a long chain, and models unloaded before a heavy pass.

• Every pass has a name you can change, and the pass types say Tile Upscale and VR2 Upscale instead of USDU.

Post FX

• Seventeen looks ship with the pack, each with a picture of what it does, and you can save your own.

• A chain list beside one editor, an Order view to arrange the chain, and a Limit on every effect: the subject, the background or a mask.

• New effects: Relight, Skin, Film Stock and Match Reference, a lens picker with ten named lenses, auto white balance, and a Colour card that is now a full grade.

Paint and previews

• Blank Canvas and Clear Canvas on the Paint tab: start from a white sheet, or wipe the lot in one press.

• Lighter previews. Live Preview and Image Review show a smaller picture on the node, which fixed the canvas lagging while you pan. Full size is one right-click away.

Fixed: a saved picture could carry the wrong prompt, Import Prompt took the wrong row, Auto Hold was slowing down runs that already fit the card, and the Overview said a SAM3 file was missing when it was there.

Removed, and why: Save Video, Video Review, the Civitai lookup and download in the LoRA panel, and the Open Folder buttons are not in this version. The Comfy Registry now scans every pack and holds back anything that runs a program or fetches files, which is why Manager was stuck giving people an old version. Taking them out is what got Manager working again. If your 1.3 workflow used Save Video or Video Review, those two nodes will show as missing. The LoRA panel still shows the Civitai links; the file is yours to fetch.

Hand written note to users

Hi, Thanks for your interest, I've been AI painting from 2023 day 1. Daily I love the process, I've also hated comfy for years, it took me thousands of hours to understand it. The worst thing about the dam software is the look, feel and experience. I wanted to PAINT, not stare at wires and understand how to hook different VAE's into different slots. I want the BULLSHIT of Comfy to be hidden behind TOOLS that do jobs you want. That was the reason I built this.

There are so many things to explain about what is here, id need a youtube series explaining each part to be honest. If you have any questions feel free to comment and ill try answer them for you. I know the rest is AI generated but its usful so please read it if you can! Thanks!

Workstudio

It brings image generation, painting, visual camera controls, references, multiple model setups, detailing, upscaling, and finishing tools together in one organized interface.

I built it for two reasons: I wanted an interface that made sense for creating art, and I was tired of the endless wires and time-consuming setup. Painting an edit or composing a shot should feel visual and hands-on. Adding another model, another refinement pass, or another stage to an image shouldn’t mean untangling connections and rebuilding half the workflow. I wanted room to experiment and build up a picture without the process becoming increasingly awkward.

Generate with one model, refine with another, paint the areas you want to change, and finish the image in the same workspace. Set up your composition on the camera stage, guide the look with a moodboard, bring in identity references, and arrange detailing and upscale passes through the panels. The included workflow connects the stages, while you decide how far to take the image.

Start with a simple generation and explore more tools as you need them, or go deeper with individual model rigs, prompts, LoRA sets, and settings for each pass. An approachable starting point for newer users, with the flexibility experienced users need to experiment.

Previously released as Krea 2 Moodboard + Identity Edit, the project has grown into RedNode Studio. Those original tools remain at its heart, now part of a broader workspace built around how you create and edit an image.

🎨 More creating. Less wiring.


Paint an edit. Set up a shot. Generate with one model, refine with another, and give the finished image your own look.

RedNode Studio brings your creative tools together in one visual ComfyUI workspace: models, prompts, references, camera controls, painting, detailing, and color grading.

I built it for painters, designers, and image makers who want more control over their pictures with less time spent navigating connections. Load the included workflow, configure your models, and work from organized panels with deeper settings available as you need them.

🎬 Direct the shot with a visual camera stage

Position your subjects, move the camera, adjust the room, and place your lights.
The stage translates your layout into camera and lighting prompts for Krea 2, with supported camera and lighting LoRAs following the controls. Explore framing, lens choices, and viewpoints or render a camera path as a sequence of still images.

🔀 Different models. Different jobs. One workspace.

Build reusable model setups called rigs, each with its own encoder, VAE, sampler settings, and LoRA set.

Use one rig to generate, another to refine, and a chosen rig for painting or detail work. Arrange refinement and detailer passes as cards, with individual prompts, denoise, scale, and settings.

🎨 Guide the style. Keep your subject in focus.

The original Krea 2 Moodboard + Identity Edit tools remain at the heart of the pack.

Use reference images to guide palette, texture, atmosphere, and subject identity. Combine moodboard styling with identity editing, adjust reference influence, and organise images into the Moodboard and the Krea 2 Identity tab, which holds Subject, Scene and Masks. More people follow the main subject as Person 2 and 3.

Identity editing uses the community Krea 2 Identity Edit LoRA.

🖼️ Rework, refine, and finish

Bring an existing picture into Img2Img, paint a local change, or explore another viewpoint with RE-ANGLE. SWAP provides reference-based face, head, or person transfer through the Qwen-Image-Edit setup.

Then build your finishing sequence: whole-image refinement, face detailing, upscaling, and Post FX. Adjust colour, clarity, bloom, depth of field, grain, and more. Follow live previews, compare previous runs, and organise your keepers inside ComfyUI.

🌱 Start simple. Go deeper when you’re ready

Honestly I can't even go into how many settings and little things have gone into creating these nodes, the experience of painting is what i was trying to capture in the node designs, there are many tiny little things that help AI artists enjoy the process instead of just staring at a bunch of wires.

• See the whole run before you start. The Overview lays the run out as boxes in the order it happens: green is on, grey is off, amber is stood aside, red wants fixing. Click a box to open its page, right-click it to switch that stage off.

• Watch your image develop. Dedicated Live Preview nodes show generation and refinement as they happen, so you can follow the process through its stages.

• Upscale with the settings already prepared. Choose from the available output resolutions using the supplied configuration, with deeper controls available when you want to customize the process.

• Build an image through progressive refinement. Start from an initial image and use the ramping scale system to increase resolution across refinement passes, developing the result in stages.

• Design your own detailing sequence. The Detailer lets you arrange passes around your process, choosing different model rigs, detailing methods, prompts, and settings for different jobs. Simple view for the main dials, Advanced for the rest.

• Paint, resize, and refine together. The Paint workspace combines masking, color guidance, editing, and resolution controls. Enlarge the working image to paint at a higher resolution, then use painting and refinement passes to develop selected areas.

• Keep using the rest of ComfyUI. Exposed outputs let you feed data into other nodes and extend an existing workflow. Use the included setup, adapt it, or make the Studio one part of something larger. Your own loaders and sampler can stand in as a rig.

• Multiple models in one workflow: keep model setups together and choose different models for generation, painting, and refinement, reducing the need for separate graphs.

• Saved prompts, and prompts from pictures: retain and reuse prompts through the built-in saved prompt system, or drop any picture on the Prompts tab and Import Prompt reads its prompt and settings back out.

• Consistent sizing and scale ratios: flexible latent sizing with a defined 1× starting point, keeping subsequent refinement and upscale sizes predictable.

• VRAM Limit: set it to your card on the Run tab and the run holds models to stay under it. The same prompt on my 32 GB card used 22.7 GB and took 1:12 uncapped, or stayed at 7.4 GB and took 1:16 capped at 8 GB. Four seconds slower, 15 GB less card.

Newer users can begin with the supplied workflow and prepared settings. Experienced users can go deeper into how each stage works and connects. The aim is to make a complex creative process easier to manage while keeping the flexibility to shape it around your own methods.

📦 Get started

The four steps are at the top of this page: install through Manager, open RedNode Start Here Install and let Manager fetch the optional packs, then open RedNode Studio. The model files are listed below, and the GitHub guide keeps the same list up to date.

The core pack adds no pip dependencies, and it downloads and installs nothing itself. It makes no network call, runs no program and reads no environment variable, and 1.4.3 passed the Comfy Registry's automated scan. Features such as SAM3 face detailing, SeedVR2 upscaling, tiled upscaling, and Qwen-based editing require their supporting models or optional packs.

Also available for Forge Neo: the Krea 2 moodboard and identity-edit toolkit. The Studio interface and workflow are ComfyUI features. Forge Neo instructions

See the repositories for full documentation, credits, and license terms.

📦 Downloads and requirements

The pack: search RedNode Studio in ComfyUI Manager, or clone the repository into custom_nodes. No pip dependencies. ComfyUI 0.26.0 or newer with native Krea 2 support, Python 3.10 or newer. The image models themselves are yours to pick. Everything past the Krea 2 block is optional: without it the pass that needed it says so and passes the picture through. Each line below is the file, the folder it goes in, and where it comes from.

Krea 2

• qwen3vl_4b_fp8_scaled.safetensors (or the bf16), in models/text_encoders, from Comfy-Org/Krea-2

• qwen_image_vae.safetensors, in models/vae, from Comfy-Org/Krea-2

• krea2_identity_edit_v1_2.safetensors, the identity edit LoRA by lbouaraba, in models/loras, from Civitai or Hugging Face.
  
  It is what the Subject tab, the face detailer and the swap presets lean on. It only behaves on the official Krea 2 Turbo: on other Krea 2 checkpoints the face drifts. Keep one rig on the official Turbo and point the face passes at it: a Detailer card names its rig, so the rest of the picture can render on any model you like and the face still lands on the official Krea2 Turbo.

Camera and light sliders

the Camera tab drives them; every slot is optional

• camera_height_krea2_rednode.safetensors, in models/loras, from Civitai or the v1.2.0 release

• camera_orbit_krea2_rednode.safetensors, in models/loras, same two places

• camera_back_krea2_rednode.safetensors, in models/loras, same two places

• Zoom slider by Loraholic, in models/loras, from Civitai

• Colour temperature slider by Loraholic, in models/loras, from Civitai

The three camera files are mine, free to use and share, not to sell. The two Loraholic sliders are theirs and only linked. The brightness slot takes whichever light slider you pick.

The Detailer

• ComfyUI-Easy-Sam3, a pack, from Manager: the face, hair and hands masks

• sam3.pt, in models/sam3, from facebook/sam3, gated: request access on the page, then download

• ComfyUI-SeedVR2_VideoUpscaler, a pack, from Manager: the upscale pass

• seedvr2_ema_7b_fp8_e4m3fn_mixed_block35_fp16.safetensors, in models/SEEDVR2, from AInVFX/SeedVR2_comfyUI

• ema_vae_fp16.safetensors, in models/SEEDVR2, from numz/SeedVR2_comfyUI

• ComfyUI_UltimateSDUpscale, a pack, from Manager: the tiled upscale pass

The SeedVR2 loaders fetch their files on first use when the folder is empty, so those two are for placing them by hand. The 3B model is smaller and the loader's default; the 7B is what the example workflow runs.

Auto prompt: Ollama and the caption engines

All optional, and nothing breaks without them: a button that needs one says so and steps aside. But this is the part people miss, so it is worth five minutes.

Ollama is the one to install first. It is a program of its own, not a node pack, and it is what Auto Sort, Rewrite and the People rewrite run on, as well as reading a picture into words. Without it those buttons do nothing.

1. Install Ollama from ollama.com and leave it running in the background. If it lives on another machine on your network, set OLLAMA_HOST.

2. Pull a model. In a terminal: ollama pull and the model's name. A vision model can both describe a picture and sort your prompt; a text-only model still does the sorting and the rewriting. Around 7 to 9 GB is the usual size.

3. Install comfyui-ollama-describer from ComfyUI Manager. RedNode talks to Ollama through the client library that pack installs, so the engine stays greyed out until it is there.

4. On the Img2Img tab's Auto Prompt page, switch Ollama on and type the model's name in the Model box.

The other caption engines are node packs, each fetching its own weights the first time it runs. Install only the ones you want; the Auto Prompt page shows what each costs before you switch it on.

• ComfyUI-WD14-Tagger, about 0.5 GB: tags rather than sentences, and the lightest of them

• ComfyUI-Florence2, about 0.6 GB: short plain descriptions, and it pairs well with Ollama sorting the result

• ComfyUI-QwenVL, about 9.4 GB

• ComfyUI-JoyCaption, about 10.8 GB: the most detailed, and the heaviest

• CLIP gen: no extra download at all

Several can run on one queue and their words join in order. Captioners are unloaded when they are done, and the Advanced tab has a button to hand back everything they are holding.

Live preview and depth

• lighttaew2_1.safetensors, the small VAE for Krea 2 live frames, in models/vae_approx, from lightx2v/Autoencoders

• comfyui_controlnet_aux, a pack, from Manager: the Post FX Depth card's estimators, each fetches its own weights on first use

Without the small VAE the live frames still stream, as the colour smear rather than a decode.

Re-angle and swap

RE-ANGLE and the multi-angle workflow run on Qwen-Image-Edit-2511; SWAP and the Detailer's swap presets run on the BFS files

• qwen_image_edit_2511_fp8mixed.safetensors, in models/diffusion_models, from Comfy-Org/Qwen-Image-Edit_ComfyUI

• qwen_2.5_vl_7b_fp8_scaled.safetensors, in models/text_encoders, from Comfy-Org/Qwen-Image_ComfyUI

• qwen_image_vae.safetensors, in models/vae, the same file as the Krea 2 one above

• qwen-image-edit-2511-multiple-angles-lora.safetensors, in models/loras, from fal/Qwen-Image-Edit-2511-Multiple-Angles-LoRA

• Qwen-Image-Edit-2511-Lightning-4steps-V1.0-bf16.safetensors, in models/loras, from lightx2v/Qwen-Image-Edit-2511-Lightning

• bfs_head_swap_v1.1_krea2.safetensors and the body swap file, in models/loras, from Alissonerdx/BFS-Best-Face-Swap

The same list, kept current, is the Downloads section of the README.