Skip to main content
Once you have an image (from a Generate → Image batch, picked in an Image Picker, or a Load Image node), the picker’s ✏️ Edit toolbar and the ComfyTV / Image menu give you a set of operations.
  • Instant (browser-side) — Crop, Rotate, Mirror, Grid Split. Parameter changes update live, output is immediately usable downstream.
  • Generative — Image Edit, Inpaint, Erase, Outpaint, Upscale, Multiangle, Relight, Cutout, plus Image Variations (multi-view / story-grid presets).
To act on a single image, work from an Image Picker — it carries the full edit toolbar.

​
Crop (instant)

Crop A draggable crop rectangle over the source image.
  • Drag the box to move it; drag the edges/corners to resize.
  • Ratio dropdown locks an aspect ratio (1:1, 16:9, …); the 🔒 button toggles the lock.
  • X / Y / W / H boxes let you type exact pixel bounds.
  • The cropped image is produced automatically.

​
Rotate (instant)

Slider for any angle (−180°…180°) plus quick buttons (⟲ 90° / 0° / 180° / ⟳ 90°). The preview rotates live.

​
Mirror (instant)

Two toggles: ⇋ horizontal flip and ⇅ vertical flip.

​
Grid Split (instant)

Grid split Slice one image into a grid → an image set.
  • Pick a preset (1×2 / 2×1 / 2×2 / 2×3 / 3×3) or use the Rows / Cols steppers.
  • Grid lines preview where the cuts land.
  • Wire the output into an Image Picker.

​
Image Edit

Generic prompt-driven image editing — change, replace, restyle, anything you can describe in words. The shipped Flux Canny Edit workflow takes an input image + a text instruction.

​
Inpaint / Erase

Mask painter A mask painter overlaid on the source image. Tools:
  • ✏️ Brush — paint the mask region (size / opacity / hardness sliders).
  • 🧽 Eraser — remove painted mask.
  • ▭ Rectangle / ◯ Ellipse — drag to draw an outline shape (annotation).
  • ① Label — click to stamp an auto-numbered marker.
  • Color applies to brush + shapes + labels; Clear wipes the canvas.
Inpaint regenerates the masked region from your prompt — ships with Flux Fill Inpaint and Fooocus SDXL Inpaint. Erase removes the masked region and fills from surrounding context, no prompt — ships with LaMa Erase (requires installing comfyui-inpaint-nodes + big-lama.pt).

​
Outpaint

Extend the canvas outward — pick the side(s) and pixels to extend, describe the whole finished image in the prompt. Ships with Flux Fill Outpaint and Fooocus SDXL Outpaint.

​
Upscale

Increase resolution. Ships with Ultrasharp 4x (4x-UltraSharp GAN).

​
Multiangle

Multiangle camera A 3D camera widget: drag the handles (or the scene) to pick a viewpoint — azimuth, elevation, distance. The chosen angle becomes a view prompt (e.g. “front-right quarter view, eye-level shot, medium shot”), then the model re-renders the subject from that viewpoint. Ships with Qwen Edit 2511 Multiangle (Qwen Image-Edit + Multiangle LoRA).

​
Relight

Clicking the Relight action spawns a pair: a pure-frontend Relight light-source node (a 3D light-ball editor — place directional / point / spot lights around a clay sphere, or click a lighting preset like three-point / Rembrandt / rim) plus an Image Stage preset to the Flux2 Klein Relight workflow (Flux-2 Klein 9B + Sun-direction LoRA). The light-ball scene renders in the browser and feeds the workflow as the lighting reference (images[1]); the model transfers that lighting onto the subject (images[0]) while preserving identity / geometry. Any image (e.g. via Load Image from Asset) can replace the light-ball render as the reference.

​
Cutout

Background removal. Wire an image, click Run, get a PNG with a real alpha channel. Ships with BiRefNet Cutout — requires birefnet.safetensors in models/background_removal/ (see models.md).

​
Image Variations

Driven by the toolbar presets — click a preset and ComfyTV auto-spawns an Image Variations stage with the right workflow pre-selected; click Run to get a set of related images. Notes:
  • Multi-view (Face / Product / Character / Multi-cam 9) — each output is the same subject from a different angle. Write the prompt as the subject description (“a young Asian businesswoman, 30s”); each branch’s angle keyword is prefixed automatically.
  • Sequence (Story 4 / Storyboard 25) — each frame is generated from the previous one (Next-Scene LoRA). Slower than multi-view.
For 360° tools see panorama.md; for picking / comparing see compose.md.