Base64ToConditioning

This is a plumbing node, not a creative one. Base64ToConditioning takes a base64 string and decodes it back into a CONDITIONING - the tensor bundle a text encoder normally produces and a sampler consumes. In other words, it's a way to carry a precomputed conditioning around as plain text and reconstitute it later, without a CLIP loader or a text-encode step in the picture.

Most people will never touch it, and that's fine. It exists for the situations where conditioning needs to travel: caching an expensive encode so you don't recompute it, moving a conditioning between graphs, or building setups where the conditioning is serialized and pasted in as a value rather than generated live. If you've ever wanted to freeze a conditioning and rehydrate it on demand, this is the "rehydrate" half.

How it works

CONDITIONING in ComfyUI is really tensors plus metadata. That can be serialized to bytes and encoded as base64 - a long, ugly, but perfectly copy-pasteable string. Base64ToConditioning does the reverse trip: it decodes the string, deserializes the tensors, and emits a normal CONDITIONING that behaves exactly like one fresh off a text encoder. The node makes no creative decisions; it's a faithful decoder. Whatever was encoded is what comes out.

The inputs and output that matter

There's really only one input:

• data (STRING, default empty) - the base64-encoded conditioning. Paste the encoded blob here. It has to be a valid encoding of an actual conditioning; arbitrary text won't decode into anything usable.

Output is a single conditioning (CONDITIONING). Wire it wherever a text encode would normally go - into a sampler's positive or negative input, or into one of RES4LYF's guide and regional-conditioning nodes.

How to install it

ComfyUI Manager: search RES4LYF, install, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/ClownsharkBatwing/RES4LYF/, activate your venv, pip install -r requirements.txt (portable ComfyUI: use the embedded python's pip), then restart and hard-refresh with F5. If you want the pack's nested sampler menu, add rgthree-comfy with "Auto Nest Subdirectories" enabled - not needed for this node specifically, but worth it for the pack.

Common issues

The whole node lives or dies on the string being valid. If the data field is empty, malformed, truncated in a copy-paste, or was encoded by something that doesn't match what this expects, you'll get a decode error rather than a picture. When it throws, check the string first: no stray whitespace or line breaks mangled in transit, and the full blob present - base64 payloads are long and easy to clip.

Also match the model. A conditioning encoded from one text encoder isn't magically portable to a different architecture - a Flux/T5 conditioning fed to an SDXL sampler won't behave. The node reconstructs exactly what was stored, so if the source and destination models disagree, that mismatch surfaces downstream, not here. If you're not doing anything that specifically produced a base64 conditioning to decode, you almost certainly want a plain CLIP Text Encode instead - this node is a niche tool for a niche need.