JoyLTX Ref Image (manual)

Refs by Name is great when you have a whole cast folder set up. Sometimes you just have one picture of one person and you want the take's character to look like them. JoyLTX RefImage is the manual version: wire a Load Image in, pick a size, and get a ready-to-use reference image for the Multishot sampler - no folders, no naming conventions, no character library. It does the same crop and sizing work as Refs by Name so the two nodes produce interchangeable output.

The required inputs are image (the reference photo), and width/height (the render size from pass 1, so the photo gets fit to it). A single image applies to every shot; a batch gives you one image per shot in order, with the last one repeating. The output trio is what the sampler expects: ref_images, ref_mask, and report. Wire ref_images and ref_mask into the sampler's slots of the same names and each shot's photo becomes an in-context keyframe at frame 0 - the person in the photo is the person in the shot, no LoRA, no captions.

Two optional inputs are where the personality lives. character is the name: fill it in and the sampler locks that character's face and voice across shots, exactly as if they were a folder character. Leave it blank and you get the photo applied without the locks - the face carries, but the voice and identity aren't tracked. And crop defaults to "portrait (keep the person, drop the set)", which takes the central upper part of the photo so the reference carries the face rather than the room. That default exists for a reason the tooltip spells out: a full-scene photo drags its set into the shot. If you want the room too, switch to "full photo" and accept the trade.

The report output tells you what it did - how many images, at what size, and whether the character is locked or unnamed. That's your check when a render doesn't look like the reference: if report says "unnamed (no locks)", the voice and identity tracking aren't active, and if you expected them, that's why.

Under the hood it crops, cover-fits to your render size, and centers on the face, matching Refs by Name's output so you can swap one for the other without retuning anything. It ships with the JoyLTX25 pack and needs nothing extra. For a single-character project, this is the node you'll actually reach for - Refs by Name is the one that scales, this is the one that's fast.