DJZ Zero EDIT Batch

DJZ Zero EDIT Batch is the sibling of the plain DJZ Zero EDIT, and it does exactly one extra thing: instead of one deterministic edit prompt, it hands you a whole list of them. Same pack, same profiles, same position-is-seed hashing. Just more.

When you'd reach for it

The single ZeroEDIT node is for when you want one specific, reproducible edit instruction. Batch is for when you want to carpet the room. If you're exploring an edit space - testing how a change plays out across many phrasings, generating a stack of variations for an A/B test, or building an automated pipeline that tries everything and keeps the keepers - this saves you from chaining a dozen single nodes together.

The honest framing: Batch is the node you use when you don't care which edit you get, you care that you get a lot of them and can reproduce the good ones. Index 17 of seed 42 is always index 17 of seed 42, forever, on any machine.

How it works

Identically to the single node, but in a loop. Each prompt is generated from (seed, start_index + i) for i in 0..batch_size-1, so a batch is a contiguous slice of index space. All the O(1) cross-platform determinism carries over - same seed and start_index deliver the same 64 prompts on a different machine later, which is what makes batch experiments actually comparable.

The inputs that matter

• profile - the same vocabulary packs as the single node, from profiles/edit/.
• seed - the world seed; different seeds, different universes.
• start_index - where the batch begins in prompt space. Think of it as the offset: shift it to pull the next chunk.
• batch_size - 1 to 64, default 4. The max is a hard cap, not a suggestion.

prefix and suffix apply to every prompt in the list, so you can stamp a common instruction onto each one. Same caveat as the single node: raw concatenation, mind your spaces.

What comes out

edit_prompts, a list of strings - and the list is the one real gotcha. You can't drag a list into a single-string prompt input. You need something downstream that consumes a list: a batch or loop node, or a string-list picker if you want to sample one at a time. The pack ships a few example workflows in examples/ that show the pattern.

Install

Same pack, same steps - it lives in the same custom node.

cd ComfyUI/custom_nodes
git clone https://github.com/MushroomFleet/ComfyUI-ZeroEDIT-nodes
pip install xxhash

Or ComfyUI Manager → search "ZeroEDIT" → Install → Restart. One dependency (xxhash), no model downloads, no GPU.

Troubleshooting

• Same profile-error trap as its sibling. A bad JSON produces [Error loading profile ...] as a list item, and your batch will happily run on it. Preview the list before you queue.
• Custom profiles only appear after a restart - the profile list is scanned when the node loads.
• "More" is not automatically "diverse." If you crank batch_size to 64, remember the default profile's arithmetic: adjacent indexes often land on the same template, so you can get dozens of near-variations. For genuinely different edits, vary the seed more than the index.

If one careful edit at a time is your workflow, the single node is the one you want. If you're screening a space of edits to find the keeper, this is the one.