Google Gemini Chat

byGoogle DeepMind

Turn moodboards, clips and brand books into shot lists, prompts and critiques, organized and ready for the next node

Break down a reference
Catalogue plate: a zinc-framed terrazzo sample board of nine swatches stamped ZANELLI LONDON, titled Terrazzo Lobby Sample Board on terracotta

How Google Gemini Chat works

Attach what you have, say what to extract, and set the format you want back. The reply arrives organized, so it can drop straight into the next node.

Fuser canvas: a text prompt and a desert house photo wired into a Google Gemini Chat node that returns a READ and a VERDICT

Attach your references

Drop in images, video frames, audio or documents, and label each one in your message so the model keeps them apart.

Google Gemini Chat node set to gemini-3.8-flash with a DP system prompt that requires a markdown shot table

Set the role and format

Give a system prompt a role and an output contract, like a markdown table or tidy JSON, then pick Flash for speed or Pro for depth.

Google Gemini Chat node output: a song brief titled Level Four with tempo, instruments, texture, opening sound and lyric

Send it down the canvas

Run the message and wire the structured reply into image, video or audio nodes as prompts, captions or shot lists.

What Google Gemini Chat is good at

Practical, inventory-style reading of images and video, camera vocabulary that holds up, and replies that obey the table, list or JSON you ask for.

Four 1970s hotel lobby photos beside an eight-shot table of framing, lens, camera movement, lighting and palette on dark bronze

Moodboard to shot list

Reads a stack of reference images and returns a structured table of framing, lens, camera movement, lighting and palette, fluent in camera vocabulary like low angle and Dutch angle.

Overhead terrazzo landing with a red marble band, above a Verona-red panel of subject, style, composition and constraints prompt slots

Reference-to-prompt drafting

Captions a reference frame in inventory-style slots (subject, medium, composition, lighting, constraints) so you can feed clean prompts to image and video nodes at volume.

Brand book digest: a large caption about a walnut Calder chair, five voice rules, and six banned words struck through in green

Whole brand book in one pass

Digests a long brand document or script in a single read and returns a taglines-and-voice summary, so campaign copy starts from the real rules instead of a skim.

Low-contrast Marlowe House lobby directory sign beside two critiques of it, one headed Flash on white and one headed Pro on black

Flash speed, Pro depth

Pick a Flash tier for brisk captioning, or gemini-3.1-pro-preview for subtler critique. The default Generic route supports attachments and all sampling controls; the alternate routed path has its own quota, so test attachments first.

Made with Google Gemini Chat

A gallery of prompts that put the model to work: shot breakdowns from video, library tagging in strict JSON, voiceover under voice rules, and a song brief from a single image.

Five frames of a sunlit concrete spiral staircase beside second-by-second production notes for camera, light and mood

Video reference analysis into production notes

Autochrome-style photo of a Viennese café mid-refit above a bordeaux panel with a six-shot table and a palette note

Role plus output contract gives a clean shot table

Four survey photos of a brutalist housing estate above JSON tags for subject, colors, lighting, visible text and search tags

Asset-library tagging in strict JSON

Voiceover script for reclaimed London stock bricks with red pause marks, ending in large type: "Ready for a second life."

Voiceover with enforced voice rules

Car park stairwell under sodium light with a painted 4, above an orange panel with the song brief "Level Four" and a lyric sketch

Image-to-song direction brief

What people build with Google Gemini Chat

Directors, prompt writers, art directors, brand teams and archivists reach for it as the eyes and organizer of a creative graph, then pass the results on to generative nodes.

Directors building shot lists

01

Hand over a moodboard or script and get a numbered shot list with framing, lens, movement and lighting you can hand to a crew or a video node.

Prompt writers feeding image models

02

Caption reference images into slot-based prompts at volume, with the target model's style and length specified in the system prompt and filler adjectives banned.

Art directors running critiques

03

Upload a layout, poster or UI screenshot and get a read on hierarchy, type and color. Step up to Pro for more interpretive, nuanced feedback.

Brand teams digesting guidelines

04

Feed in a long brand book or script, then pull out voice rules, taglines and product stories that hold to them across a whole campaign.

Archivists tagging media libraries

05

Caption stills, clips and audio into consistent descriptions, alt text and tags. Use the lighter Flash-Lite entries for batch jobs and turn on Last Message Only so each file stays independent.

Google Gemini Chat specs

Plans and credits
Versions
gemini-3.8-flash, gemini-3.7-flash, gemini-3.6-flash, gemini-3.5-flash, gemini-3.5-flash-lite, gemini-3.1-flash-lite, gemini-3.1-pro-preview, gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-lite
Inputs
Prompt (text), System Prompt (text)
Output
text, image

Gemini Chat questions

Yes. Gemini reads video and audio as well as images, so attach the clip to the Attachment input and ask for a numbered shot list with framing, camera move, lighting and approximate timing. Asking for one shot per line makes it easy to split the list and feed each shot to a video node.

Yes. By default each run sends the conversation history along with your new prompt, which helps when you're refining one idea over several passes. For batch work like captioning a set of frames, turn on Last Message Only so earlier answers don't leak into the next one. Rules that should apply every time belong in the System Prompt.

Start with the default gemini-3.8-flash for most prompt drafting and captioning. Drop to the flash-lite entries for batch captioning of many frames, and choose gemini-3.1-pro-preview for long multi-document synthesis or subtle visual critique. Pro is slower and costlier, so don't spend it on simple captions.

The node accepts text, images, video and audio, plus documents. Say explicitly what to extract, such as palette, lens, lighting, on-screen text or pacing. With several attachments, label them in the prompt, like 'Image 1 is the style ref', so they stay separate.

Not reliably. The image output is conditional and only fills when the selected model returns an image, so most text-tier selections leave it empty. Treat it as a text collaborator that writes prompts, shot lists and critiques for your image nodes.

No, not guaranteed. Sampled outputs vary. Reusing the same seed, prompt and model improves repeatability, but it is best-effort. Keep temperature low for captions and structured output, and pin a model for repeatable pipelines, since preview models can change over time.

Try Google Gemini Chat on Fuser

Turn moodboards, clips and brand books into shot lists, prompts and critiques, organized and ready for the next node

Break down a reference

Related models

Google Gemini Chat docsAll models

OpenRouter Chat

OpenRouter (aggregator); underlying models vary by selection

Perplexity Chat

Perplexity AI

ChatGPT

OpenAI

Groq Chat

OpenAI

Mistral Chat

Mistral AI

Claude Chat

Anthropic