IPAdapter Slider

Style transfer from reference images is the IP-Adapter trick, and on Flux it's called Redux. IPAdapter Slider is the Flux Continuum pack's control for up to three Redux reference images at once: three strength sliders - IP1, IP2, IP3 - packed into a single VEC3 output. Set how much each reference should influence the generation without hunting through a stack of IPAdapter nodes.

How it works

Each of the three inputs is a strength value, 0 to 1, default 0. At 0 the corresponding reference has no influence; at 1 it's pushed hard. The node bundles the three floats into a VEC3 tuple, and the workflow splits it back out (via SplitVec3) to feed the individual strength inputs of the Redux/IPAdapter applicator for each reference image.

The naming is a small history lesson: this slider is called "IPAdapter" because that's the family of techniques it controls, but the pack's own README renamed the module to Redux to match BFL's terminology - the Redux model (flux1-redux-dev.safetensors plus the sigclip vision model) is what actually powers image-variation conditioning on Flux here. In the Redux world, strengths are intentionally low: the README's advice for Redux references is "use very low strength values," because Redux is strong and 0.3 can already dominate an output. Each Redux input maps to one reference image (Redux 1, 2, 3).

Inputs and outputs

• IP1 - FLOAT, 0.0–1.0, default 0.0, step 0.05. Strength of the first Redux reference.
• IP2 - FLOAT, 0.0–1.0, default 0.0, step 0.05.
• IP3 - FLOAT, 0.0–1.0, default 0.0, step 0.05.

Output is one VEC3 (IP1, IP2, IP3).

Installing it

Part of robertvoy/ComfyUI-Flux-Continuum:

cd ComfyUI/custom_nodes
git clone https://github.com/robertvoy/ComfyUI-Flux-Continuum

or ComfyUI Manager ("Flux Continuum"), restart. To make it do anything, you need the model files: flux1-redux-dev.safetensors in models/style_models/ and sigclip_vision_patch14_384.safetensors in models/clip_vision/, from the README's model list.

Where people get burned

The defaults are all 0 - the sliders do nothing until you raise them, which is correct but easy to miss ("why isn't my reference showing?" - because IP1 is 0). And the "very low strength" advice is not optional: on Redux, moderate values (0.5+) can flatten the whole image toward the reference's composition, overriding your prompt. Start at 0.1–0.2 and nudge up. If you only use one reference, just leave IP2/IP3 at 0 and ignore them.