Rebel VAE Encode 🗜️

The KSampler doesn't work on images - it works on latents, the compressed latent-space representation that's all a diffusion model actually manipulates. Rebel VAE Encode 🗜️ is the step that gets your starting image into that latent space: feed it pixels plus a VAE and it returns a LATENT. This is the node you reach for in img2img work, inpainting setups, or any workflow that starts from an existing image rather than an empty latent. The Rebel version is the pack's animated stand-in for ComfyUI's VAEEncode - same behavior, glowing box.

How it works

Standard pack architecture: grab the native VAEEncode input definitions, append theme and color_scheme, delegate the encode call to core ComfyUI. The pixel-to-latent conversion is untouched - you get exactly what the stock node would produce. The animation is a frontend overlay drawn behind the widgets, with zero effect on the encode and no VRAM cost of its own.

The inputs that matter

• pixels - the IMAGE to encode, wired from your Load Image node or anything upstream that produces pixels.
• vae - the VAE to encode with, from your VAE loader. The encode/decode pair should generally use the same VAE.
• theme / color_scheme - the 12 animations and 14 palettes. Decoration.

The output is a single LATENT, which feeds your KSampler's latent_image input - typically with a denoise below 1.0 if you want to keep the image's structure rather than regenerate from scratch.

Installing it

One clone, no dependencies:

cd ComfyUI/custom_nodes
git clone https://github.com/RealRebelAI/Rebels_Animated_Nodes.git

Restart ComfyUI, then hard refresh the browser (Ctrl + F5 / Cmd + Shift + R) so the cached JavaScript reloads and the animation draws. It's under Rebel AI → Animated Core.

Where people get burned

• "The result ignores my input image." If your KSampler's denoise is at 1.0, the encoded latent is basically discarded - full regeneration. Lower denoise (say 0.4–0.7) to actually lean on the encoded image.
• Wrong VAE = garbage latent. Encode and decode with a consistent VAE family. Feeding a mismatched VAE in gives you a latent that decodes into chaos.
• No animation after install. Browser cache. Hard refresh - it's the pack's #1 support issue - and re-refresh after pack updates.
• Flat on Nodes 2.0. The pack draws on the legacy LiteGraph canvas; the encode works but the backdrop can vanish under the newer frontend.

Simple node, simple job. If your workflows start from real images and you want the whole chain to match, this is the entry point to the Rebel look - and if the aesthetic doesn't matter, the stock VAEEncode is identical underneath.