🖼️ Gemini Session Gallery

The most underrated node in the ComfyUI-Gemini-Conversation-Canvas pack. 🖼️ Gemini Session Gallery does something no other node here does: it hands you every turn of a conversation at once, as one batch image plus a text log of what happened at each step. When you've chained five edits and can't remember whether "add rain" came before or after "make it evening," this is the node that answers it.

How it works

Give it a session - from Session Start, Edit Turn, or Load - and it walks through the session's saved turn images, loads each PNG from the session folder, and stacks them into a single IMAGE batch tensor. That batch is exactly what a Preview Image or Save Image node wants, so one wire gives you the full before/after progression side by side in the UI.

Two implementation details are worth knowing. First, because you can change aspect ratio between turns, the images don't always share dimensions - the node pads them all to the largest size with black bars so they fit in one batch. A 16:9 turn next to a 1:1 turn will show up letterboxed, which is the batch tensor being honest rather than cropping. Second, alongside the batch it builds a turn_log: a plain-text summary where each line is Turn N: followed by that turn's model reply, truncated to 120 characters. It's the quickest way to reconstruct what was actually said.

Inputs and outputs

One input - session - and three outputs:

• images - the batch of all turn images, top of the stack being the most recent. Wire into Preview Image to see the progression, or Save Image to export the whole sequence as numbered frames.
• turn_log - the per-turn text summary. Wire into a text display node, or just glance at it in the UI.
• turn_count - how many turns the session has, useful for sanity checks or feeding into a comparison node.

Note that Gallery doesn't change the session or make any API calls - it's pure local reading of what's already saved. There's no extra billing on this node, which makes it a safe thing to sprinkle through a workflow.

Install and gotchas

Same pack, same install - ComfyUI Manager (search "Gemini Conversation Canvas") or:

cd ComfyUI/custom_nodes/
git clone https://github.com/jeremieLouvaert/ComfyUI-Gemini-Conversation-Canvas
pip install -r ComfyUI-Gemini-Conversation-Canvas/requirements.txt

Restart. Dependencies: google-genai, filelock, Pillow - no model downloads, and as noted, no API calls on this node.

The gotcha is subtle and mostly harmless: the gallery reads the session's saved images, so it reflects whatever has actually been written to disk - which is every turn, since each is saved the moment it's generated. And if a session has no images at all (say, a freshly created session that never generated anything), you get a 64×64 black placeholder and a (no images) turn log rather than a crash. One genuinely useful pattern: run [Session Load] → [Session Gallery] → [Preview Image] any time to get a contact sheet of a long-running edit project before you decide what to change next.