RedNode Studio - ComfyUI for Artists | Paint, Camera & Multi-Model Workflows - v1.4 Showcase
Every picture in this article was made inside one node. No wall of wires, no hunting for the group somebody forgot to un-bypass. I build RedNode Studio because I was tired of opening a workflow and needing a map to use it. This is a walk through the panel as it stands in 1.4.2, page by page, with what each part is for.
⚡ 60 nodes, no pip dependencies. Search RedNode Studio in ComfyUI Manager, install, restart. The source and the full README are on GitHub.
How the showcase pictures were made
Krea 2 Turbo at 8 steps on the Workspace's own sampler. Then the Detailer: a tiled upscale, a SeedVR2 pass to 4K, and two face passes, one on Krea 2 and one on Z Turbo. Post FX on top for the grade and the grain. That whole chain is one Generate press. The prompt and the settings are on each picture, so you can read exactly what I used.
The Overview: the whole run at a glance
This is the page I open first. The run is laid out as boxes in the order it happens: what feeds the render on top, then the canvas, the passes, the Detailer, Post FX and Save. Green is on, grey is off, amber is on but stood aside, red wants fixing, and hovering a box says why. Click a box and it opens the page that decides it. Right-click a box and it turns that stage on or off, which is new in 1.4.2 and the thing I now use most. Workspace Presets sit at the top, so a whole setup loads in one pick. At the bottom it lists every node pack and model these settings need, ticked if you have them. It installs nothing itself. ComfyUI Manager does that, where you can see it.
Models: rigs, not loaders
A rig is a model with its CLIP and VAE, its sampler numbers and its Detailer steps, saved under a name. Keep several and switch with a click. A Detailer pass can name any of them, which is how one run uses two models without a second graph. The seed for the whole run lives here too.
Prompts: built box by box
Style, Subject, Surroundings, Light And Colour, each in its own box with presets, and a live preview of the paragraph they make. Type a messy prompt into Anything Else and Auto Sort files it into the right boxes. Rewrite gives you a fresh variation that keeps the subject and moves the rest. Import Prompt reads the prompt out of any picture you drop on the page, and after a run the tab shows the exact words that were sent, wildcards rolled and all.
🎬 Camera: a stage seen from above
Place the subject, move the camera, set the height and the lens. It writes the framing as the physical-camera language Krea 2 actually follows, a lens, a height, a distance, an angle, worked out from where things stand. Fifteen sets ship. A camera path renders one picture per shot in a single queue, and the lights write their own paragraph.
LoRAs: a stack you can read
Rows drag by their grip, switch off by their eye, and group under titles that fold and switch off together. Named sets live on their own tabs, and a rig, a Detailer pass or a paint pass picks its set by name.
Latent: the canvas, and passes on top of it
Pick a shape or drag the canvas, and it tells you what that size costs in VRAM before you press anything. Then up to ten passes, each with its own denoise and scale, with a Ramp that sets them all in one click. Draft small, climb in steps, no second node.
Img2Img: a source, its passes, and the words for it
Drop a picture in or wire one. Each pass can run on a different rig, so one pass can be handed to another model. Auto Prompt describes the picture with any of six caption engines and puts the words where you tell it: into a prompt row, before your words or after them. Re-angle re-shoots the source from another viewpoint and Swap puts a face, a head or a whole person onto it.
🎨 Paint: fix the part, keep the rest
Mask a region or paint colour straight on the picture, set the denoise, Generate. It composites back by itself and the result sits underneath, ready to save or paint on again. Auto Mask finds the subject or the background for you. New in 1.4.2: Blank Canvas gives you a white sheet at your Latent size to paint on from nothing, and Clear Canvas takes the picture off as well as the paint.
👤 Krea 2 Identity: keep the face
Drop a face into Subject and Krea 2 keeps the identity across renders. More people follow as Person 2 and 3, Scene rebuilds a place, and Masks say where the identity lands. This is the part of the pack this listing started with, and it is folded into the Workspace so nothing needs wiring.
Post FX: the grade, in camera order
Seventeen looks ship and you can save your own. Underneath, every effect has its own dials: denoise, colour, clarity, sharpen, relight, haze, lens distortion, depth of field, chromatic aberration, bloom, light wrap, diffusion, vignette and grain. The Order view sets the chain.
The Detailer: passes you read top to bottom
The post-render work is a list of cards with no wires between them: a sampler pass, a face detailer, a SeedVR2 upscale, a tiled upscale, in any order, in groups you can fold and switch off together. This page used to look like a cockpit, so 1.4.2 adds a Simple view: the rig, the target, the size, the strength bars and the references, and nothing else. Advanced shows every setting. Switching views changes no value, and a card carries a small chip when a hidden setting is in use, so nothing runs behind your back. Every pass now has a name. Left alone it names itself, Tiled Upscale, Face Detailer 2, and you can type your own.
The VRAM Limit
Set it to your card and the run holds models to stay under it. The same prompt on a 32 GB card: uncapped it used 22.7 GB and took 1:12. Capped at 8 GB it stayed at 7.4 GB and took 1:16. Four seconds slower, 15 GB less card. A rig that would not fit your card renders on it anyway. The Run tab shows every stage with its time, the live picture, and the VRAM as a chart while it goes.
New in 1.4.2
Workspace Presets on the Overview, with the panel as it stands reading Current.
Right-click a stage on the Overview to turn it on or off.
Detailer Simple view, named passes, and pass kinds that read Tile Upscale and VR2 Upscale.
Blank Canvas and Clear Canvas in Paint.
Fixes: a saved picture could carry the wrong prompt when a workflow had a second sampler chain, Import Prompt took the wrong row, and Auto Hold was slowing down runs that already fit the card.
Install and first run
ComfyUI Manager, search RedNode Studio, install, restart. Open the template browser and load RedNode Start Here Install first: it holds one node from each optional pack, so Manager offers them all in one pass. Restart, then load the RedNode Studio template. A note inside it lists every model and LoRA file it uses and where to get each one.
The Krea 2 nodes need ComfyUI 0.26.0 or newer. On anything older the other fifty nodes still load and work with any model. The optional packs are only needed for the feature that uses them, and the Overview tells you which.
The published pack makes no network call, runs no program and reads no environment variable.
If you try it and something is confusing, tell me. The confusing parts are the ones I fix first.