Create Media

Media nodes run generative media models. Use them to generate, edit, and convert between images, videos, 3D models, and audio. Each media node is titled with the name of its model. To add one, click + next to the agent composer and choose a tile under Quick create: Create image, Edit image, Replace region or Make video. Pick an Output quality (Draft, Balanced or Quality) first to choose a faster or higher-fidelity model. For any other media model, click Browse all to open the node picker. The node generates one output per input, like every other node in Runchat. Provide a list of prompts and you’ll get a list of outputs in parallel. When a node has several outputs, a variation strip shows of and lets you page through them.

​

Selecting a model

Select the node and click the model button on its toolbar (Swap model) to open Choose a model and pick from any installed media model. Media nodes only show models that produce media (images, video, 3D, audio). Pricing for each model is visible in the picker. You can swap models without losing your prompt or other inputs, useful for trying the same prompt across several models. See Models for installation and Credits for the full pricing table.

​

Prompt

Use the prompt input to describe what you want the output to look like, or what you want changed when editing an existing image. Most current models (Nano Banana, Gemini Image, Flux Kontext, GPT-Image) accept instruction-style prompts like “change the shoe to red” or “add fog to the background”, and follow them closely. A few older image-to-image models like SDXL behave differently: they treat the prompt as a description of the whole image rather than an instruction. If a model isn’t responding to instructions, this is usually why.

​

Reference inputs

Some models take a reference image or video as input: image-to-image, image-to-3D, and image-to-video models, for example. When you select one of these, the node adds the necessary input automatically. Wire in any image or video output from another node, drop a file onto the input row, or click Add to upload one.

​

Mask

Some models accept a mask to specify the region of the image to edit. The quickest way to set one up is + > Replace region. On a mask input, click Draw to paint the mask over the source image in the focused view.

​

Toolbar controls

Select a media node to show controls relevant to the chosen model:

• Model (Swap model): pick a media model from your installed library
• Variations: how many variants to generate per run (1 image, 2 images…)
• Edit: opens the image in the focused view. See Image editing.
• More (…):
  • Size: for image and video models, set the aspect ratio or resolution
  • Duration: for video models, select the output length
  • All parameters: every model parameter, each with Show as input to expose it as a handle
  • Run details: the model, duration and Credits used for the last run
  • Download / Download all: save the generated files at full resolution

​

Image editing

Click Edit on the node toolbar, or switch the View toggle at the top of the editor from Workflow to Image, to open the image in the focused view. Mark areas, add reference images and describe a change in the Edit image panel, then click Apply. Each edit is saved as a new version you can restore from History. See Image Editor.

​

Using LoRAs

To generate with a LoRA you trained:

1. Install a LoRA-capable model from Fal: open the node picker, click Add Fal, and paste the model URL (for example a Flux 2 LoRA endpoint).
2. On the dashboard Training page, click Copy model URL on your LoRA.
3. Ask the agent to apply the LoRA URL to the node at a given scale, for example: “use this LoRA on the selected node at scale 0.8”.

​

Next steps

• Models: install additional models from Fal or OpenRouter
• Credits: cost per image/video/3D model
• Custom LoRA models: train and use your own LoRAs
• Generate images with reference images: worked example using image-to-image