Questions and answers

The easiest way is ComfyUI Easy-Install: download it from its GitHub page, make a new folder (not straight in C:\ itself), extract the zip there and double-click ComfyUI-Easy-Install.bat. It installs Git, Python, ComfyUI and the Pixaroma nodes by itself. No Git on your PC yet (or an old one)? Windows asks if it may make changes: press Yes.

You need Git installed. Open your custom_nodes folder (portable: ComfyUI_windows_portable/ComfyUI/custom_nodes), type cmd in the address bar, and run git clone https://gitlab.com/pixaroma/comfyui-pixaroma.git. Then restart ComfyUI and press Ctrl Shift R.

Installing from the Manager also works, but it only offers the old version 1.4.144 (see the flagged question below).

Close ComfyUI, open the ComfyUI EZi Launcher and press Easy-Install under UPDATE: it updates Easy-Install itself and gets the newest Pixaroma nodes. To update ComfyUI too, open the launcher again and press ComfyUI Stable. Then start ComfyUI and press Ctrl Shift R in the browser.

Open the ComfyUI-Pixaroma folder inside custom_nodes, type cmd in the address bar, then run git pull. Restart ComfyUI and press Ctrl Shift R in the browser.

Close ComfyUI first. With Easy-Install, press ComfyUI Stable under UPDATE in the EZi Launcher. With the portable, run update_comfyui_stable.bat from its update folder. Installed with git: git pull in the ComfyUI folder, then install what it needs (the help page has both commands). Then start ComfyUI again and press Ctrl Shift R.

Both. Some parts of the nodes only load when ComfyUI starts, so restart it first. When it is running again, press Ctrl Shift R once in the browser so the new screen files load too.

Click the ? button in ComfyUI's top bar. Your version is in the bottom right corner of the Pixaroma Help window; clicking it copies the full details. Press Latest version next to it to see the newest version and what changed in it: if they match, you are up to date.

You can also add the Version Check Pixaroma node: it shows your Pixaroma, ComfyUI and frontend versions, with a Copy button for bug reports.

The update did not land, and clearing the browser cache will not help. If you updated with the Manager, it only offers 1.4.144: switch Pixaroma to nightly in the Manager, or update with Easy-Install or git pull instead.

Still the same? Reinstall cleanly: close ComfyUI, delete the ComfyUI-Pixaroma folder in custom_nodes (delete it, do not just rename it, or ComfyUI loads both copies), clone it again from GitLab, start ComfyUI and press Ctrl Shift R. Your fonts and presets live outside that folder; only sounds you added to Notify are inside it, so copy those out first.

If you installed Pixaroma through the Manager, its folder is not a git copy, so git pull quietly updates something else and never touches Pixaroma. Switch Pixaroma to nightly in the Manager, or delete the Pixaroma folder and clone it again. Also check that only ONE Pixaroma folder is in custom_nodes, and that you fully restarted ComfyUI.

The GitHub copy is kept level with GitLab, so it still gets the current version. GitLab is the main home, so it is better to repoint once: in the ComfyUI-Pixaroma folder run git remote set-url origin https://gitlab.com/pixaroma/ComfyUI-Pixaroma.git.

No, it is only a label. The Manager recognises installed packs by their address, and Easy-Install gets Pixaroma from GitLab, which the Manager does not match. Your nodes and updates are fine, so you can ignore the button.

Yes, since version 1.4.170. Older versions showed empty Pixaroma nodes and runs failed with missing ResolutionState under a path prefix: update Pixaroma. With nginx, also pass the Host header through (proxy_set_header Host $host), or ComfyUI refuses the script files.

Almost never. The browser still runs the old node files from its cache. Click on the ComfyUI page and press Ctrl Shift R once (on a Mac Cmd Shift R): the page reloads with the new files and the buttons come back.

Still wrong? Press F12, open the Network tab, tick Disable cache, keep the tools open and reload the page.

Add a Version Check Pixaroma node. If its Pixaroma line turns orange with a warning, the browser runs older files than the ones installed: press Ctrl Shift R. If the versions match, it is something else: ask in #pixaroma-nodes on Discord and say which node.

It is probably switched off, not broken. Open Settings, find the Pixaroma section and tick the Show the ... button in the top toolbar box you want back. Hiding a button never turns the feature off: Alt W still opens Workflows and Alt H still opens Help.

ComfyUI has two looks for the same nodes: the classic look used in the videos, and the newer Nodes 2.0. To switch, open the ComfyUI menu (the logo at the top left) and turn Nodes 2.0 on or off. The Pixaroma nodes and their settings are the same in both.

Fixed in version 1.4.158: update Pixaroma. If it ever guesses your theme wrong, set Node look in the Pixaroma settings to Dark (or Light) instead of Auto.

Fixed in version 1.4.207: update Pixaroma. Titles now follow the node colour in the light palette in both node looks, also inside an opened subgraph and on the Show advanced inputs tab in Nodes 2.0.

Look at the red node's name. If it ends in Pixaroma, it is newer than your Pixaroma nodes: update, restart ComfyUI, and press Ctrl Shift R.

If it comes from another pack, open the Manager, click Install Missing Custom Nodes, press Install next to the pack, then Select. When it is done press Restart, confirm, and press Ctrl Shift R.

Some packs need extra files that did not install. Look at the black console window of ComfyUI while it starts: the red error text there names what is missing. Copy that text when you ask for help.

First check that the update really took: click the ? in the top bar and compare your version with Latest version. If it did not go up, or it did and the node is still red, reinstall the Pixaroma nodes cleanly, then press Ctrl Shift R.

No. When the nodes are missing completely (not just blank), the pack failed to load when ComfyUI started, and a browser refresh cannot help. Fully restart ComfyUI and look in the black console window for red text near the word Pixaroma. Post that text on Discord with your versions.

Since version 1.4.148 the pack lists the video helper it needs, so Easy-Install and Manager updates install it: update Pixaroma. After a git pull, install it once yourself. Portable: open a command window in the folder that holds python_embeded and run python_embeded\python.exe -m pip install imageio-ffmpeg. With your own Python, run pip install imageio-ffmpeg in that environment. Then restart ComfyUI.

This happens with a workflow made on another computer: it points to a folder on THEIR PC. Click Browse on the node and pick a folder of your own once; after that it always works. ComfyUI's own input, output and temp folders never need approval.

Update Pixaroma. Music Prompt crashing ComfyUI 0.37 was fixed in version 1.4.159, and AI Prompt or Music Prompt on a list (for example several pictures from Load Images from Folder) was fixed in version 1.4.203. ComfyUI's own Generate Text node can still crash on a list.

Fixed in version 1.4.198: update Pixaroma. Those models make pictures with a transparency channel, which the older text nodes read wrongly. Now a picture with transparency stays transparent, and normal pictures are unchanged.

Fixed in version 1.4.175: update Pixaroma. Before that, ComfyUI's node refresh (the R key) turned Set and Get into empty stand-ins; reload the page with F5 after pressing R until you update. Workflows saved while it was broken repair themselves when opened.

In Pixaroma workflows, press the Download Models button on the canvas: it lists every model with its download link and the exact folder inside ComfyUI/models (diffusion_models, text_encoders, vae, loras...). Put the file there, then click the empty canvas and press R, or restart ComfyUI, so the lists see it. Other people's workflows usually have the links in a note on the canvas.

Quick way: unzip it and drag a .json file onto the ComfyUI canvas. Tidy way: copy the whole episode folder into your workflows folder (Easy-Install: ComfyUI-Easy-Install/ComfyUI/user/default/workflows), open the Workflows panel on the left, press its refresh button and click the workflow.

The Ming Image workflows need ComfyUI 0.38.0 or newer. Older ComfyUI does not know the Ming text encoder and stops with this error before any Pixaroma node runs. Update ComfyUI, restart, and check that Load CLIP shows the Ming text encoder with type qwen_image.

That is a ComfyUI core node, not a Pixaroma one, and older ComfyUI does not have it. Update ComfyUI (the Qwen Image 2.1 workflows work on 0.37.0 or newer), restart, and press Ctrl Shift R.

The 27B prompt writer is very tight on a 16 GB card and can write broken text, which Ming then draws. Switch the prompt writer loader to Qwen3.5 9B and pick the matching Ming preset in the AI Prompt node. The writer runs on your own PC, so a VPN does not matter.

Put the .safetensors file in ComfyUI/models/background_removal/. Standard (birefnet.safetensors, 424 MB, 4 to 6 GB VRAM) suits clean objects and products. High Resolution (birefnet-hr.safetensors, 8 GB+) keeps fine outline detail. Matting (birefnet-matting.safetensors, 8 GB+) is best for hair, fur and soft fabric. With no model, the node tells you what to download and where.

Those two download from HuggingFace as model.safetensors and must be renamed to birefnet-hr.safetensors or birefnet-matting.safetensors. The name decides the working size (2048 for hr or matting, 1024 for any other name), so a wrongly named HR file loses the point of HR. On Windows, tick File name extensions on the View tab first, so the rename keeps .safetensors.

No, it is optional. The Remove Background node uses the BiRefNet files only. rembg only adds extra choices to the AI background removal buttons in Image Composer and Paint. To add it on Easy-Install, run python.exe -m pip install rembg in the python_embeded folder, then restart ComfyUI; its models download on first use.

Double-click an empty spot on the canvas and type Pixaroma: every Pixaroma node shows up. Click one, then click the canvas to place it. You can also click Nodes on the left and open the Pixaroma folder, where they are sorted into folders like Editors, Image and Values.

That is how newer ComfyUI places a node: move to where you want it and click the canvas once.

Click the ? button in the top toolbar (or press Alt H). The Help window covers every Pixaroma node, the canvas tools and short guides, and its search reads the whole text, so you can type a problem like "buttons missing". You can drag a node card from it straight onto the canvas.

It opens Pixaroma Workflows (also Alt W), a browser for the workflow files you already have: nothing is imported or moved. Each workflow gets a picture, and the search reads inside the files, so you can type a model name, a LoRA or words from a prompt.

Click the mountain icon in the top toolbar, next to the Manager. While it is on, dragging or resizing a node snaps its edges and centers to nearby nodes, with thin orange guide lines. Hold Shift to skip snapping for one drag. The snap distance is under Settings, Pixaroma, Align (advanced).

Put .ttf or .otf files in ComfyUI/models/fonts, open a font list and click the ↻ button next to its filter box: your font appears under Custom, no restart needed. Text Overlay, Text Watermark and the text layers of Image Composer all use this list. For real Bold and Italic, copy in every file of the family (Regular, Bold, Italic...).

Text Overlay Pixaroma draws a styled caption; Open Text Editor gives a full-screen canvas to drag, scale and rotate it, and you can wire text into it. Text Watermark Pixaroma has no editor: pick one of nine anchor points plus a margin, and size the text in pixels or as a percentage of the image width, so mixed sizes stay consistent.

Each node has an open button: Open Paint Studio, Open Image Composer or Open 3D Builder. For Paint and Image Composer you can also drag a picture onto the closed node to add it as a layer. Work in the full-screen editor, click Save, then Run the workflow to get the picture out.

Use Save Image Pixaroma: type or paste a folder path, or click Browse to pick one. Leave it empty for the output folder. The Will save as line shows the exact file the next Run creates; files are never overwritten, and a / in the name makes subfolders.

The Browse window is the older Windows one, but you only need it once per top folder: everything inside a picked folder is approved too, so after that you can type any path below it.

PNG is lossless, keeps transparency and the workflow, so you can drag it back into ComfyUI. WebP is usually the best middle road: much smaller, keeps transparency and still reloads the workflow. JPG makes small files but has no transparency, and only has room for about 64 KB of workflow (since 1.4.177 it keeps at least the prompt). Test a file by dragging it back into ComfyUI: Windows Properties does not show this data.

Yes, any setting of any node. Type %NodeTitle.setting% with the title shown on the node, for example img_%KSampler.sampler_name%_%KSampler.steps%_%counter%. The Will save as line shows the result straight away. With two samplers, give them different titles.

Turn on Add Civitai generation info in the node's settings (the gear, or right-click). It is on Save Image, Preview Image, Save Mp4 and Save Video. For pictures Civitai then fills in the model, every LoRA with its strength, steps, seed, sampler and size; videos show the prompt and settings. If Civitai showed nothing on older saves, update: fixed in version 1.4.199, so save new copies to upload.

Save Disk saves the shown picture to a folder you choose, Save Output writes it to ComfyUI's output folder, Copy puts it on your clipboard and Open shows it in a new tab. Click a thumbnail to enlarge it and use the arrow keys to flip through a batch. Set save_mode to save to write every frame to output on each Run.

Save Mp4 is the quick one: drop it in and press Run. Use Save Video when you want your own folder, file name tokens, or to see the exact file name first; it can also write MP4 HQ (H.265, 10-bit) for smoother gradients. Both need an even frame width and height, and say so clearly if they are odd.

Use Load Video Pixaroma: upload a file or pick one, and it plays on the node. Wire its video_frames and audio outputs straight into Save Mp4 Pixaroma. Max frames, Force FPS, Skip first frames and the custom size control what you load. A "video reader" message means the av package is missing: install it with the Manager's pip option.

Load Image Pixaroma adds a resize panel (Max megapixels, Longest side, Fit inside, Crop to fill, Pad and more). Load Image Mini is the compact one with the resize options in its gear, and outputs only image and image_info; add Image Info Pixaroma to unpack the mask, width, height and file name. In either picker, click + Folder to add a picture folder from anywhere on your PC.

Fixed in version 1.4.177: update Pixaroma. Ctrl V now pastes pictures the normal way, with no permission popup, and copied nodes paste again. The Paste button on Load Image Mini still has to ask, so use Ctrl+V with the node selected.

Wire Seed Pixaroma into KSampler's seed input. Random rolls a new seed every Run, Fixed keeps the same one for repeatable results, and New fixed random rolls one and locks it, so you can keep a lucky result while changing other things. Use last seed brings back the previous Run's seed (after you ran once in this session).

No LoRA loader can send words down the CLIP wire: it carries the model, not text. Use the loader's triggers output and wire it into a Text Join node's text box; the words follow the LoRAs you switched on. Click the i on a LoRA row to see exactly which words are sent.

Add Run Timer Pixaroma anywhere on the canvas, it needs no wires. Right-click it and choose Run Timer settings, turn on Chime on finish (this timer), pick a Sound and the Volume, and press Preview to hear it. It plays when the whole Run is finished. If no timer ever chimes, check the mute button in those settings: it mutes every Run Timer.

Wire your inputs into the rows of Switch Pixaroma (up to 32, any type) and click a row to make it active (orange). Only the active row passes through, and only its branch runs. Double-click a row name to rename it; disconnect a wire to remove its row.

Put your nodes between Loop Start and Loop End, wire Loop Start's loop output into Loop End's loop input, and set total to the number of rounds. To keep each round's result, feed it into a value slot on Loop End and use Combine Pixaroma inside the loop to collect them. The index output counts the rounds from 0.

Yes, with Pause Image Pixaroma. Put it between the picture and the slow step, set it to Pause and Run. Like it? Switch to Continue and Run again: only the steps after it run, fed from the picture you saw. Regenerate makes a fresh one, Pass runs everything in one go.

Put Notify Pixaroma at the end of the workflow, pick a sound and volume, and use Preview to hear it. For your own sounds, put .mp3, .wav or .ogg files in the assets/sounds folder inside the Pixaroma folder and restart ComfyUI. It also times how long the run took to reach it.

Monitor counts memory that ComfyUI keeps parked for reuse as free, because ComfyUI can use it again straight away; Task Manager counts it as used. The Free VRAM button on Monitor gives that memory back, and then the numbers come closer. The pale mark on the VRAM bar is the highest point of the last run.

Put Free VRAM Pixaroma on the wire between the two stages, so the first model is let go before the second loads. Only its input matters: drag the wire from what you want cleaned up after onto it; the output does not need connecting. All also hands spare memory back to the card, Models drops only the models, Cache keeps the models loaded.

No, it runs on your own PC with a model in your ComfyUI/models/text_encoders folder, with no account or key. You write the instruction (the formula) once in its gear settings. With no model chosen, it passes your text through unchanged. Video Prompt Pixaroma works the same way, with a vision model of roughly 5 to 10 GB.

The enhancer workflows use temperature 0.3, which keeps the text close to your idea: change a few words, or raise the temperature in the node's gear settings (for example to 0.7). Since version 1.4.179 it has extra picture inputs, image_2 up to image_8.

Both are fixed: undo and redo keeping every Pixaroma node's live controls in version 1.4.186, and workflows opening as modified on ComfyUI frontend 1.53 in version 1.4.197. Update Pixaroma and refresh the page. A redo (Ctrl Y) that disappears after a click is a ComfyUI bug, fixed in ComfyUI frontend 1.55.11: update ComfyUI for that one.

Fixed in version 1.4.153: update Pixaroma. The prompt and text boxes now let ComfyUI's shortcut through, so a selected word becomes something like (cat:1.1). The same update fixed middle-mouse panning over Pixaroma nodes in the classic look.

Load Images from Folder passes the whole folder as one batch, not one Run per picture. Update Pixaroma (fixed in version 1.4.146). For one result per line use Prompt Each; version 1.4.203 also added a One file per entry option to Save Text.

They did, and both are fixed: Group Switch in version 1.4.158 and Run Timer (with 2 or 3 decimals) in version 1.4.163. Update Pixaroma.

The editor only keeps what you painted when you press Save. Since version 1.4.112, closing it without saving asks first. Update Pixaroma, and press Save before you close the editor.

Fixed in version 1.4.201: update Pixaroma. Before that, Prompt Reader only read PNG files.

The Comfy Registry marks our versions after 1.4.144 as flagged after an automatic review, with no reason given, so the Manager only offers 1.4.144 and misses every newer node. The nodes are safe and work normally. To get the newest version, update with Easy-Install or git pull (both get it straight from GitLab), or switch Pixaroma to nightly in the Manager and restart.

Yes. The pack is not banned: its status on the Comfy Registry is Active, and an earlier ban was lifted on 11 September 2026. You can check it yourself on the registry's status page.

Post in #pixaroma-nodes on the Pixaroma Discord (ComfyUI questions not about a Pixaroma node go in #comfyui). Send your versions (the Copy button on Version Check Pixaroma), say what happened and what you expected, add a screenshot, and copy any red error text from the black console window. Try Ctrl Shift R first: it fixes most "it looks broken" reports.

No, issues and pull requests are closed. Bugs, questions and ideas all go to #pixaroma-nodes on the Pixaroma Discord, which is free to join.

Yes, free, under the MIT license: you can use, copy, change and share them, as long as the copyright notice stays with them.