Generate Image (from 10 Images)

DeepGen_I2IX is the pack's maximum-context image node: up to ten reference images feeding one generation. Where DeepGen_I2I3 gives you enough references to pin down a single subject, ten is the tier where you're defining an entire visual world - a character from multiple angles and their wardrobe and the setting and an object they always carry, all reconciled into one output. The "X" in the name is the pack's way of marking it current; the deprecated spelling that old workflows use is DeepGen_I2I10.

When ten references earns its keep

This is the node for anything where the model needs to hold a lot of detail that prompting can't convey. A fully-specified character with consistent costume pieces. A product plus its packaging plus reference shots of the environment it sits in. A scene where three or four characters each need identity. The reference budget isn't there to be filled - it's there so you can throw everything at the model and let it figure out the visual rules. Models in the dropdown are the ones with the context windows to use it: GPT Image 1.5, Seedream 5.0 Fast Lite, and the two Nano Bananas (Pro and 2). If a model you'd expect isn't listed, it doesn't support this many references - the dropdown only shows what can actually take them.

The inputs

• image_1 through image_10 (IMAGE) - all optional, all wired the same way. Use one or use all ten; unconnected sockets are ignored. Source filenames get traced through your graph so each attachment arrives on DeepGen's side labeled by what it was.
• prompt - the instruction that tells the model what to do with all that context. The more specific you are about which reference is which ("the person is from images 1–3, the jacket from image 5"), the less the model has to guess.
• model, minimum_resolution, aspect_ratio, output_format, nb_results, seed_value, output_prefix, config_json - the standard DeepGen set; see the DeepGen_T2I0 article for what each does.

Outputs

The usual triple: IMAGE (the result, ready for preview or save), output_prefix_and_model (STRING for naming), and total_credits_used (FLOAT - and with this much input, worth watching).

Install

Same pack, same routine:

cd ComfyUI/custom_nodes
git clone https://github.com/deepiksdev/ComfyUI-DeepGen-API.git
cd ComfyUI-DeepGen-API
pip install -r requirements.txt

Restart, then paste your DeepGen API key (deepgen.app) into Settings → "DeepGen API Key"; it lives in ComfyUI/user/deepgen/config.json.

Gotchas

This is the most expensive node in the image family per call, so use nb_results sparingly. Uploading ten images also means ten images of your material leaving the machine - fine for fictional characters, less fine for anything sensitive. And when results disappoint, the first thing to try isn't more references, it's a sharper prompt that labels what each reference is for; these models read ten images as context, not as a checklist. If you spot the old name DeepGen_I2I10 in a legacy workflow, it's the same node - new work should use DeepGen_I2IX.