Reve Image Remix
Mash up to six reference images into something new
- reference_images
- IMAGE
This is the most interesting node in the Reve family. Instead of a single source image, it takes up to six reference images and blends them under a text prompt - your moodboard as input, a new image as output. If you've ever wanted to say "combine the lighting from this shot, the pose from this one, and the color palette from that one," this is the node built for exactly that.
The reference system is the standout feature: the prompt can point at specific images by index using XML tags. Write <img>0</img> to reference the first image, <img>1</img> for the second, and so on. So you're not just blending - you can instruct "use the subject from <img>0</img> with the environment of <img>1</img>." That level of control over a multi-reference generation is rare, and it's why Remix is the node Reve fans actually talk about.
Inputs that matter
- reference_images - an autogrow input, 1 to 6 IMAGE inputs, each with its own wire
- prompt - multiline, up to 2560 characters, and the place to use your
<img>N</img>tags - model -
reve-remix@20250915(quality) orreve-remix-fast@20251030(quick and cheap), plus aspect_ratio and test_time_scaling (1–5) sub-options - upscale - disabled or 2x/3x/4x, adds cost
- remove_background - boolean, adds cost
- seed - re-run control, non-deterministic results
Output is a standard IMAGE tensor.
How it works
Each reference image is base64-encoded and sent with your prompt to /proxy/reve/v1/image/remix; the result comes back as a WebP binary, decoded into a tensor. You need at least one reference image - the node errors out if you try to run it empty - and no more than six. Hosted and billed per call from your Comfy account credits. Ships with ComfyUI core, no install, needs a Comfy account.
Common issues
- The tags matter more than the prompt. Without
<img>tags, the model blends all references loosely and "who's the subject" gets ambiguous fast. Explicit tags are the difference between a deliberate composition and a soup. - Six references is a lot to steer. It works, but each extra image dilutes the others. For most remixes, two or three references with clear roles beat a six-image pile.
- Content policy, again. A flagged result raises an error server-side; no way around it.
- Deprecated. The Reve family was flagged for deprecation in August 2026 (PR #15331). Functional today, but this integration is being wound down - treat it as a tool, not a platform.
Reach for Remix when you're doing character/style mashups, product concept boards, or anything where "blend these references this way" is the actual job. It's the one Reve node with no obvious local equivalent.
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| reference_images | COMFY_AUTOGROW_V3 | — | |
| prompt | STRING | Text description of the desired image. May include XML img tags to reference specific images by index, e.g. <img>0</img>, <img>1</img>, etc. | |
| model | COMBO | Model version to use for remixing. | |
| upscale | COMBO | Upscale the generated image. May add additional cost. | |
| remove_background | BOOLEAN | false | Remove the background from the generated image. May add additional cost. |
| seed | INT | 00–2147483647 | Seed controls whether the node should re-run; results are non-deterministic regardless of seed. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |