Claude Chat

Name: Claude Chat
Author: Anthropic

byAnthropic

A literate creative collaborator that holds one voice, reads your reference stills, and turns loose briefs into shot lists and prompts

Hand over the brief

Brass paraffin lighthouse lamp with a tall glass chimney on grey, on a pale blue catalogue page titled Carrow Head Lighthouse Lamp

How Claude Chat works

Set the tier and voice once, attach what it needs to see, and take back a draft already shaped for the next node in your graph.

Claude Chat node settings: model claude-opus-5-5, a voiceover system prompt, and the slash-marked voiceover it returned

Pick a tier and a voice

Choose a Haiku, Sonnet, Opus, or Fable-class model, then put your tone, banned phrases, and output format in the system prompt.

Text prompt and a perfume reference still wired into a Claude Chat node that returns a READ note and numbered shot list

Hand over the material

Write the creative job, audience, and length, and attach labeled stills, layouts, or text briefs for it to read first.

Claude Chat node output: a READ description of a perfume still followed by shots 1 to 4 with lens, move, light and action

Pipe the draft downstream

Ask for a fixed structure, such as numbered shots or labeled sections, and feed the result straight into the next node on your canvas.

What Claude Chat is good at

Sustained prose, literal-then-interpretive image reading, and film and design vocabulary make it the thinking layer that sits upstream of your pixels, sound, and motion.

Pink script page: "No vessels." three times in grey, then "One vessel. Pedestrian." in bold Courier above the lantern-room scene

One voice across long scripts

Holds a single, even-tempered voice through treatments, scripts, and long-form copy. Put a short sample of the target voice in the system prompt and it shifts easily: terse, lyrical, or deadpan.

Perfume bottle on wet travertine beside a green panel with its READ description and six numbered shots in gold

Shot lists from reference stills

Reads an attached still literally first (layout, color, subject), then interprets it into numbered shots with lens, camera move, and lighting. Sonnet-class handles everyday boards; Opus-class catches finer visual detail.

Two full-bleed photos of a black rubber glove holding a red tulip over ice, each above the model-specific prompt that made it

Prompts tuned per target model

Writes separate, model-specific prompts for your image and video nodes, one paragraph per variant. Use the default route for standard behavior with attachments and system prompts; the alternate route is a fallback worth testing before you rely on it.

Film poster with three numbered rings on black beside "Remove the cobalt entirely." in large type, three fixes and a verdict below

Staged design critique

Reviews a screenshot or layout in sequential passes (hierarchy, type, color, copy) and returns concrete fixes, not vague verdicts. Move to an Opus-class tier when the critique needs subtle visual judgment.

Made with Claude Chat

The gallery shows the kinds of text it hands back: voice-shifted copy, art direction briefs, asset tags, sonic direction, and spoken lines.

Three amber taglines with deadpan, warm and lyrical copy for a beeswax candle, beside a tall crop of the candle and a match

Campaign copy: one product, three brand voices

1930s diving-tower photo above three colour bands, Pool Cobalt, Albumen Cream and Warm Concrete, each with its line from the brief

Moodboard to art direction brief

Close crop of a crackled celadon tea bowl above a celadon panel with its alt text set large and five catalogue tags

Image description: alt text and asset tags

Diner guest check: "Last Pie Under Glass", a 52-bar strip, and a huge mustard 41 beside the note on where the bass drops out

Sonic direction for a late-night track

Riso poster: "We don’t make promises." in black, a pink slash, and "We make it roll." on a pink field above the full voiceover

Voiceover with marked pauses

What people build with Claude Chat

Writers, directors, art directors, and prompt builders reach for it when a project needs one steady voice or a careful read of a brief, a still, or a layout.

Screenwriters and directors

Draft treatments, scripts, and dialogue that keep one voice from first page to last, then rewrite scenes in a different register without losing the thread.

Cinematographers and previs artists

Turn a brief or a reference still into a numbered shot list with lens, move, and lighting, ready to pin beside storyboards.

AI artists building prompt chains

Use it as the prompt-writing layer between a loose idea and your image or video nodes: name the target model, set length limits in the system prompt, and get clean variants back.

Art directors and designers

Get a second pair of eyes on posters, layouts, and UI screenshots, with feedback sorted into hierarchy, type, and color passes instead of a single blur of opinion.

Brand copywriters and strategists

Shift taglines, product stories, and social captions between brand voices, or digest a long brand guide into a single page of rules the whole team can follow.

Claude Chat specs

Plans and credits

Versions: claude-sonnet-5-5, claude-opus-5-5, claude-fable-5-1, claude-sonnet-5, claude-opus-5, claude-fable-5, claude-opus-4-8, claude-opus-4-7, claude-opus-4-6, claude-opus-4-5, claude-sonnet-4-6, claude-haiku-4-5
Inputs: Prompt (text), System Prompt (text)
Output: text

Claude Chat questions

Yes. Every current Claude model accepts image input, so connect your stills to the Attachment input and ask for what you need, such as lighting, lens, palette or wardrobe. With several images, label them in the prompt ("Image 1 is the style reference") so details don't blend. Wire the Last Response output into an image node's prompt to use the result directly.

Yes, when the voice is spelled out. Put the voice rules and a short sample passage in the System Prompt, and leave Last Message Only off so each reply can see earlier scenes. Sonnet 5.5 (the default), Opus 5.5 and Fable 5.1 have a 1M-token context window, enough for a full script plus a style guide; Haiku 4.5 has 200K.

Start with a Sonnet-class model, the balanced default, for drafting, prompt-writing, and image reading. Move to Opus-class for subtle critique, dense documents, or long scripts that must stay coherent. Fable-class sits above Opus for long, ambiguous work. Haiku suits fast, high-volume drafts like captions and tags.

No. It accepts text and image-style attachments and returns text only. It can't analyze video or audio files, and it won't produce pixels, sound, or motion. What it can do is write the shot lists, critiques, and model-specific prompts that your image, video, and audio nodes work from.

Not reliably. Claude has no seed setting, and current models like the default Sonnet 5.5 run with fixed sampling, so Temperature, Top P and Top K don't apply. For the most repeatable results, put your rules and one example output in the System Prompt, turn on Last Message Only so earlier turns don't shift the reply, and keep the prompt and model the same between runs.

Claude is made by Anthropic. The earliest release date recorded for models in this node is 2025-10-15. Newer tiers are reported to offer much larger context windows, but check Anthropic's docs for current limits.