RedNode Studio
A studio rig for ComfyUI: a tabbed workspace panel, painting with auto masks, wireless control panels, LoRA management, grading and prompt tooling, plus Krea 2 moodboard and identity editing.
Nodes (63)
Adding a third reference to a Krea 2 edit (and when not to)
Keep the face, change everything else
Turn reference images into vibe, not content
Moodboard Encode
Krea 2 Moodboard + Identity Fusion
The 12-layer tap that decides your detail
Krea 2 RedNode, moodboard + identity in one node
Load a saved Krea 2 look like a config file, not a lucky guess
Freeze a Krea 2 look you love, before the next run erases it
Every Krea 2 RedNode dial, named in plain English
Camera sliders that follow your stage, so zoom and height stay in sync by themselves
Re-shoot an existing photo from another viewpoint, driven by the same stage you planned the original with
A camera you drag on a map, instead of words you hope the model obeys
The type converter you didn't know you needed, for when a string must become a number
Drive another node's dropdown without a single wire
Every dropdown on your graph, on one node, with not a wire in sight
Turn a reference picture into a full prompt, section by section, with a local model
The node that unloads your first model so the second one actually fits
Switch whole workflow groups on and off from one panel, then save the combinations
One dropdown that decides which groups run and which get bypassed
A prototype that tells you which groups will run before you waste a queue on it
The preview that doesn't forget your last ten runs
Watch the render form instead of squinting at a colour smear
The LoRA stack that knows your trigger words and rolls random strengths
Save the LoRA stack you just dialed in, before you lose it
A glowing signpost for your workflow, not another widget to wire
Restyle every note on your graph from one place, no hunting
The node that stitches your painted fix back in, feather and all
Paint with any renderer you like, not just the one that ships with the tab
Renders only the region you painted — the rest of the image never moves
Route your whole graph by color, not by a wall of boolean gates
A wire that flips your routing colors as the data goes through it
Grade images in physical camera order, with zero extra diffusion passes
The Workspace's grading chain, pulled out into your graph
The prompt box that highlights tokens and rolls its own wildcards
Gluing prompts together without rewiring
A framing dial that stops your subject dissolving into the wide shot
The shared @keyword library every RedNode prompt box reads
A caption converter that swaps her→his without mangling 'there'
Cut a face out at real resolution and refine it with any sampler
Put the refined crop back so only the masked region changed
A graph sign that tells you what value just ran through it
The front door your external engine's render walks back in through
The prompt and sampler numbers your own KSampler never got handed
Load the model your way, still get the Workspace's LoRAs and prompts
Hand NovelAI its prompt, seed and sampler numbers without touching its UI
The node at the end of your own rig, with nothing on the right of it
A colour-driven switch where the losing branches never even run
One switchboard for every colour-driven Router in your workflow
Stops you re-typing steps and CFG every time you switch checkpoints
A save node that files your images and writes the prompt down beside them
A dropdown of your own choices that outputs the picked string
A Load Image that finally gives pictures back
Photograph any point in the graph so you can compare runs later
Compare any two points of the last run with a drag-across wipe
The post-render passes as a visual list — sampler refines and face detailers with zero wires between them
A local LLM that reads your settings and leaves them alone
The whole studio input rig in one tabbed panel, wired by a single bundle
Pull values out of a subgraph by name, with no wires through the boundary
Push values into a subgraph by name, spliced into real wires before the run
The switch that doesn't run what you didn't choose
Assembling prompts from parts with RedNode Text Combine
RedNode Wildcards just writes a plain .txt — which is why it's useful
RedNode Studio
One panel for every model and every job. Render, edit, paint, upscale and finish a picture from a single node in ComfyUI, on Krea 2, Qwen Image 2.1, Z-Image, Anima, SDXL, Illustrious or Pony.
<table> <tr> <td width="33%"><img src="images/run-tab.webp" alt="A Krea 2 render on the Run page, with the pipeline, the finished picture and the VRAM chart"></td> <td width="33%"><img src="images/identity-run.webp" alt="The same woman in a new scene, rendered from a Hero Creator headshot"></td> <td width="33%"><img src="images/run-rail-right.webp" alt="An Anima render with the tab rail on the right"></td> </tr> <tr> <td align="center">Krea 2, from a two line prompt</td> <td align="center">Same face, new scene, from one photo</td> <td align="center">Anima, or any model you like</td> </tr> </table>63 nodes · No pip dependencies · Every model, every purpose
Watch the full tutorial: install, models and every tab you need to make, fix and save a picture.
Quick install on Windows
Two files on the latest release set everything up, on the portable ComfyUI or the desktop app:
- Download RedNode_Install_Node_Packs.bat and RedNode_Download_Models.bat.
- Put both in your ComfyUI folder: next to
run_nvidia_gpu.batfor the portable, or in the desktop app's base folder, the one withcustom_nodesandmodels. Run from anywhere else, they find the desktop app's folder and ask before doing anything. - Close ComfyUI and run RedNode_Install_Node_Packs.bat. It installs RedNode Studio and the node packs its workflows use through ComfyUI Manager, and skips any you already have.
- Run RedNode_Download_Models.bat and type the numbers you want: the models, LoRAs and tools, each from its publisher into the right folder. Press Enter for the recommended start, or A for everything, which is well over 100 GB.
- Start ComfyUI, open Templates, then RedNode Studio, and load RedNodeStudio_V1.6.
Both files are plain text, so you can open them in Notepad and read every line before you run them.
Any other install
Search RedNode Studio in ComfyUI Manager, or clone it:
git clone https://github.com/RedNodeAI/ComfyUI-RedNodeStudio.git ComfyUI/custom_nodes/ComfyUI-RedNodeStudio
Then open RedNode Start Here Install from the template browser, let Manager install what it offers, restart, and open RedNode Studio. The full guide covers every tab, node and setting.
Why I built it
A ComfyUI graph for a finished picture turns into a wall of wires: loaders, samplers, detailers, upscalers, grading, saving. I wanted one place where all of that lives, where switching models is a click, and where the canvas stays small enough to see. The RedNode Studio Workspace is that place. You pick a model, write a prompt, press Generate, and the render, the detail pass, the look and the save all happen inside one panel.
Every model, set up in one click
Press the family you have and the rig is built: the model, its text encoder, its VAE and the settings that model wants, found in your own folders by name. Krea 2 Turbo, Qwen Image 2.1, Z-Image Turbo, Anima and SDXL / Illustrious / Pony are ready to go, and any other model loads by hand. Keep as many rigs as you like and switch between them in a click.
A studio, not a graph
<table> <tr> <td width="50%"><img src="images/overview.webp" alt="The Workspace Manager: the whole run in order, and what needs attention"></td> <td width="50%"><img src="images/prompts.webp" alt="The Prompts tab with a Frame box"></td> </tr> <tr> <td><b>See the whole run at a glance.</b> Every stage in order, green for on, grey for off, and a plain word on anything that needs fixing. Click a box to jump to it.</td> <td><b>Write prompts that build themselves.</b> Style, subject, surroundings and light in their own boxes, a camera that writes its own sentence, wildcards and saved presets.</td> </tr> <tr> <td><img src="images/editor-realism.webp" alt="The Re-render tab"></td> <td><img src="images/paint.webp" alt="The Paint tab"></td> </tr> <tr> <td><b>Edit any picture.</b> Turn an illustration into a photograph, re-shoot it from a new angle, swap a face in, upscale it, all from the Tools tab and Re-render, on a gallery picture or on the render you just made.</td> <td><b>Paint what you want changed.</b> Mask a region, type what should be there, press Generate. Only the painted part changes, and the result becomes the next thing you paint on.</td> </tr> <tr> <td><img src="images/post-fx.webp" alt="Post FX with seventeen shipped looks"></td> <td><img src="images/camera.webp" alt="The Camera tab"></td> </tr> <tr> <td><b>Finish it like a photographer.</b> Twenty effects in camera order, from exposure and skin to film stocks and lens glass, and seventeen ready made looks to start from.</td> <td><b>Direct the camera.</b> Place the subject and the camera on a stage, pick the lens and the height, and the prompt and the camera LoRAs follow.</td> </tr> </table>One photo in, the same person anywhere
<table> <tr> <td width="50%"><img src="images/hero-headshot.webp" alt="The Hero Creator: source photo, clean crop and front on headshot"></td> <td width="50%"><img src="images/identity-run.webp" alt="The same woman on a beach at sunset"></td> </tr> </table>The Hero Creator takes any photo, cuts the head out above the clothing and renders a clean front on headshot. Drop that into the Subject gallery and Krea 2 Identity puts the same person in any scene you write: a hotel lobby in green velvet becomes a beach at sunset in white linen.
Built to run on your card
Watch every stage with its time, the picture forming step by step, and a VRAM chart that shows what is on the card. Set a VRAM limit and the Workspace works out what the run needs and holds it under your card's size, so a big model still runs on a small GPU.
Keep what you make
File every picture by date, preset, seed, model or size, keep drafts apart from keepers, and cull a session without leaving ComfyUI. Every file carries its prompt, its settings and the model and LoRA hashes, so a picture posted to Civitai lists what made it.
Make it yours
Put the tab rail on either side, fold it to icons, hide the tabs you never use, or pick a preset for the job: Basic, New image, Image to image, Edit, Paint or Finishing. A red Generate sits at the top wherever you are.
And the rest
A Detailer that lists its passes instead of wiring them, a LoRA stack with search and groups, moodboards, image to image with several passes, a local assistant that explains any setting, seeds you can link across the whole run, live previews, a picture history with rerun, stage by stage comparisons, and control panels that drive other nodes with no wires at all.
Read the full guide for every tab, node and setting, the model downloads, and performance numbers. What changed in each version is in the changelog.
License
PolyForm Noncommercial 1.0.0. Free for personal use, hobby projects, research, study and noncommercial organizations. Commercial use needs my permission, so get in touch. Full terms in LICENSE.
In plain language: use it, modify it, learn from it, share it, all fine. Selling it or building it into something you sell is not, unless you have arranged that with me separately.
ComfyUI itself is GPL-3.0 and is a separate work. Releases up to and including v1.4 were published under GPL-3.0 and stay available under those terms. The license change applies from v2.0 onward.
Credits
ComfyUI. lbouaraba/ComfyUI-Krea2Edit, Apache-2.0, the identity-edit dual-conditioning recipe this reimplements. Jonseed/ComfyUI-Detail-Daemon, MIT, the detail curve the rig's Detail Daemon dial re-implements. nova452/ComfyUI-ConditioningKrea2Rebalance and huwhitememes/comfyui-krea2-conditioning, Apache-2.0, the per-layer rebalance mechanic and its RMS-renormalized variant. no8d/ComfyUI-NO8D-controls, MIT, the LoRA-stack row-list UI that inspired RedNode LoRA Stack, implemented independently here. skatardude10/ComfyUI-Optical-Realism, which I read as a survey of which optical effects were worth having while building the grading chain. Loraholic for the Krea 2 zoom and colour temperature sliders the Camera LoRAs card drives. ethanfel and ostris for the Krea 2 vision-conditioning recipes. yijunwang2 for the krea2-anypaint LoRA and its inpaint recipe, which the Paint tab's AnyPaint follows. Krea.ai for Krea 2, under the Krea Community License.
Not affiliated with Krea.ai.