Depikt — Curated GPT Image 2 Prompt Library + Generator + Gallery + Critique

Source: depikt.app landing + depikt.app/library URL: https://depikt.app/

Depikt is a web app at depikt.app billed as a “curated prompt library for GPT Image 2.” It mixes a hand-picked prompt catalog (~350 prompts at fetch) with three ancillary surfaces — Generate (crafts custom prompts from rough ideas), Gallery (user outputs), and Critique (feedback). Each library card pairs a preview image with the full prompt text, revealed on click. Lighter and more editorial than the open-source awesome-gpt-image-2 catalog (4,400+ prompts), and structurally a workflow surface rather than a flat reference list.

Key Takeaways

• GPT Image 2 specific. Depikt’s marketing names GPT Image 2 explicitly. No mention of GPT Image 1.5, DALL-E 3, or other image models — the catalog is tuned to GPT Image 2’s capabilities (text rendering, character consistency, commercial-quality illustration).
• 350+ curated prompts in a browsable grid with thumbnail previews. Smaller and more editorial than awesome-gpt-image-2 — fewer prompts, tighter selection.
• Eleven category filters: All, Posters, Infographics, UI Mockups, Social Posts, Cinematic, Storyboards, Interior / Food / Fashion, Visual Summaries, Image Edits, Open-Ended Creative. The taxonomy skews toward design / marketing output types — useful for the WEO Marketly visual-content workflow.
• Click-through reveals the full prompt text. Each card unfurls into preview image + complete prompt — meaning Depikt is also a prompt-engineering reference, not just a thumbnail gallery. You can study or copy the prompt source verbatim.
• Generate crafts a structured GPT Image 2 prompt from a rough natural-language brief. Positioned as the path when no curated prompt fits — sits next to the Library, not behind it.
• Critique is a feedback surface — specific functionality not detailed in public copy. Worth opening to confirm whether it’s iterative-feedback-on-output or peer-critique.
• Gallery surfaces community outputs from prompts run on Depikt — the visual proof-of-capability layer. Good for spotting what GPT Image 2 actually delivers vs. the claim.
• Pricing / sign-up / maker info NOT disclosed on the public landing. No tier pricing, no team page, no API documentation surfaced via WebFetch. This is the article’s load-bearing open question.
• Editorial selection vs community firehose. Depikt is hand-picked; awesome-gpt-image-2 is community-sourced from public Twitter / 𝕏. Use Depikt when you want a vetted starting point; use awesome-gpt-image-2 when you want breadth or a specific style track.

Example prompt titles seen on the library

(For taste-calibration — which output types Depikt’s curators prioritize.)

• TIME Magazine Cover Recreation
• Deconstructed Multimedia Portrait Collage
• Street Dance Choreography Sheet
• UK Citizenship Identity Card
• Luxury Skincare Campaign Ad

The range is intentional: editorial mimicry, fine-art collage, instructional sport, official-document mimicry, brand campaign. Depikt’s library is a working “show and tell” of GPT Image 2’s range, not a single-style track.

Implementation

Tool/Service: Depikt (web app at depikt.app)

Setup: None — browse depikt.app directly. No install, no CLI, no agent integration documented on the public landing.

Cost: Not disclosed on the public landing. Probable model: free browsing of the Library + paywall on Generator / Critique (typical for prompt-library web apps). Verify via sign-up flow before assuming free access.

Integration notes:

• Depikt outputs prompt text — to actually generate the image, paste the prompt into ChatGPT, OpenAI Platform’s image API, the Codex CLI image_generation tool (see Codex CLI’s built-in image_generation tool over ChatGPT OAuth (zero-incremental-cost on ChatGPT Plus / Pro; gpt-image-1.5 + gpt-image-2 reachable via codex exec against the ChatGPT image quota)), Higgsfield’s GPT-Image-2 backend (see Higgsfield Overview), or any other surface that drives GPT Image 2.
• The “Generate” tool is a prompt-construction helper, not a hosted image generator (so far as the public landing shows). The image generation itself happens elsewhere — Depikt prepares the prompt; you bring the model.

Try It

• Find a starter prompt for the WEO Marketly weekly infographic. Filter Library → Infographics or Visual Summaries. Pick one whose composition matches the data shape (radial / grid / timeline). Copy the prompt verbatim, swap WEO topic + data, paste into the Codex CLI’s built-in image_generation tool over ChatGPT OAuth or Higgsfield’s GPT-Image-2 backend. Compare output quality vs the current Higgsfield-only WEO infographic pipeline.
• Reference the UI Mockups category for product-screenshot prompts. When the WEO product / sales team needs a hero mockup of a dashboard or app screen for marketing collateral, Depikt’s UI Mockups filter is the highest-density starting point — the prompts already encode the composition discipline that distinguishes a real-looking mockup from generic AI slop.
• Study a prompt unfurl as prompt-engineering reference. Open a Cinematic-category card, read the full prompt, note the structural elements — composition language, lighting language, lens / shot specification, color palette specification, post-processing language. Export the structural template as a reusable WEO chatgpt-image-prompt-template.md like the existing marketing-infographic template in ~/.claude/brand-assets/weo-marketly/.
• Use Generate to convert a Mel feedback note into a generated visual. When the Mel feedback (internal editorial-rules doc)|Mel feedback (internal editorial-rules doc) flow surfaces “I don’t like this image” without a clear alternative, drop the rough idea into Depikt’s Generator and use the structured prompt as the starting point for a regeneration cycle.
• Cross-check against awesome-gpt-image-2 before committing. Depikt’s 350-prompt curation may miss what’s hot in the larger 4,400-prompt corpus. For any high-stakes deliverable, search the awesome-gpt-image-2 corpus too — broader coverage, daily updates, more language coverage.

Related

• awesome-gpt-image-2 — YouMind-OpenLab’s 4,400-prompt GitHub catalog — the larger, open-source, community-sourced sister resource. Depikt’s Library is the curated minority report; awesome-gpt-image-2 is the firehose.
• Codex CLI’s built-in image_generation tool over ChatGPT OAuth (zero-incremental-cost on ChatGPT Plus / Pro; gpt-image-1.5 + gpt-image-2 reachable via codex exec against the ChatGPT image quota) — how to run a Depikt-sourced prompt against GPT Image 2 from Claude Code without a Platform API key.
• Higgsfield Overview — Higgsfield’s product-photoshoot backend uses GPT Image 2 (via FAL); Depikt-sourced prompts are drop-in compatible.
• Higgsfield + Claude Code Ad Agency Workflow — applied DTC ad-creative pipeline that’s Depikt’s natural downstream for marketing-output prompts.
• AI Web Design — UI Mockups category overlaps with this topic’s spec-prompt aesthetic; for full landing-page generation see ai-web-design/.
• Prompt Engineering — Depikt’s click-through prompt unfurls are useful reference material for building reusable image-prompt templates.

Open Questions

• Pricing tier and sign-up flow. Not stated on the landing or library page. Free / freemium / paid? Whether Generator / Critique are paywalled. Worth a sign-up before recommending Depikt as a workflow surface for the WEO team.
• Maker / team / company. No About page surfaced via WebFetch — who runs Depikt, where they’re based, funding posture, longevity outlook. Important for any tool the WEO team would route into a recurring workflow.
• Critique feature scope. “Critique” appears in nav but functionality not described in public copy. Iterative feedback on a generated output? Peer critique on prompts? Distinguishes Depikt from a flat prompt list and is worth verifying.
• API surface. No documented API. Without one, Depikt is a manual browse-and-copy surface — not a Claude-Code-or-skill-integration target. Would change the recommendation if an API exists.
• Update cadence. Repo-style libraries (awesome-gpt-image-2) advertise daily updates; Depikt’s update cadence is undisclosed. Curated catalogs can stale fast.
• Output-licensing model. Who owns prompts copy-pasted out of Depikt? “Curated” implies the team has done some editorial work; the licensing of the curated text matters for any commercial WEO deliverable.