AI Chat Tools in Photography: Composition and Post-Processing Advice
Vendor guidance warns that AI-generated photography advice can be unreliable, so composition and post-processing suggestions need visual verification.
Sources checked 2 Oct 2026
For AI chat tools in photography, OpenAI’s File Uploads FAQ, Claude’s file-upload guide, and Gemini’s file-upload help document image inputs, but do not establish which assistant gives better composition or post-processing advice. Chat Picker has not tested ChatGPT, Claude, or Gemini for this use; its method relies on published vendor information. This rewrite removes the older page’s unsourced statistics and test results and replaces them with an afternoon test plan.
Know these limits before you start
As read on those pages in October 2026, OpenAI says images are capped at 20 MB each and Free users at 3 uploads a day. Claude lists a 500 MB chat-upload limit per file, while its vision documentation allows 20 images per Claude.ai message. Gemini supports photos and up to 10 supported files per prompt, subject to availability. Check your current account limits before testing.
Before uploading client images, compare data settings. OpenAI says an opt-out is available on Free, Go, Plus, and Pro. Claude says it is available on Free, Pro, and Max, while Team is not trained on by default. The Gemini plans page states no training setting, so Chat Picker marks that cell “Not verified.”
In the same October 2026 read, OpenAI warns that ChatGPT can be incorrect or misleading and sound confident when wrong. Claude gives the same warning about occasional incorrect or misleading responses. Google says language models are probabilistic and may reflect training-data limits. None of those pages benchmarks photography advice, so compare visible claims with the image. If your test includes realistic changes to an identifiable person, OpenAI’s policy prohibits using someone’s photorealistic image without consent in ways that could confuse authenticity.
The test plan
Use the same image you have permission to use across all three assistants. Record refusals, missing features, and unusual behavior as part of the test rather than treating them as failures.
- Composition and rule-of-thirds accuracy
Prepare: A street portrait with a face near an edge, a lamp post, a visible horizon, clipped highlights, and blocked shadows. Keep a grid overlay beside the original.
Give: “Analyze this street portrait for composition. Identify the rule-of-thirds intersections, the horizon, the subject’s visual weight, and distracting elements. Separate visible facts from suggested crops, and for each crop name any edge object that could be lost.”
Look for: Claims that match the overlay and visible scene, plus crops that account for the face, lamp, and horizon.
Log: Mark each observation as supported, uncertain, or contradicted by the photograph. Record any crop constraint the assistant overlooked.
- Post-processing workflow accuracy
Prepare: An unedited copy and a short statement that the intended output is for the web.
Give: “Create a non-destructive post-processing workflow for a web version of this image. Order the exposure, contrast, color, local-adjustment, sharpening, and output steps. State the purpose of each step, tie it to visible evidence, and flag every setting you cannot infer from the file.”
Look for: A usable sequence, reasons tied to the image, and clear distinctions between observation and assumption.
Log: Note missing stages, conflicting instructions, invented settings, and any step you could not perform in your editor.
- Speed, depth, and visible tools
Prepare: The same photograph in a fresh chat. Record the clock time before submitting the prompt.
Give: “Propose three ways to strengthen this composition while preserving the subject’s expression. Explain the visual trade-off in each proposal and identify the exact region that changes.”
Look for: Distinct alternatives, concrete visual reasoning, and any source or tool labels shown by the product.
Log: Record elapsed time, how many suggestions remain usable, and which controls or source features appeared. Do not infer hidden processing that the interface does not show.
- Context retention
Prepare: Continue the conversation from the composition test without re-describing the image.
Give: “Without restating the image, list the constraints from your previous answer. Then revise the strongest crop so it keeps the face, lamp, and full horizon visible, and identify anything you cannot verify from this conversation.”
Look for: Retention of earlier constraints and honest acknowledgment of missing information.
Log: Record contradictions, repeated questions, or new assumptions. Keep these separate from errors in the first response.
- From chat to edit
Prepare: The agreed edit plan and a list of controls available in your photo editor.
Give: “Turn the agreed edit plan into a non-destructive checklist for a raw-image editor. Separate changes made before export from export-only changes, and mark every instruction that depends on camera, lens, color profile, or file metadata.”
Look for: Ordered actions you can perform without guessing and clear handling of missing metadata.
Log: Note handoff gaps, irreversible steps, and unsupported claims. Treat invented camera settings or visual details as reasons not to use the response for a final edit.
How to read your results
Before testing, write your own must-have conditions: visual accuracy, a usable crop, an executable workflow, retained constraints, acceptable elapsed time, and an acceptable data setting.
Review each log in the same order. Count a response as supported only when its visual claims match the photograph. Treat missing information as missing rather than filling the gap yourself, and do not reward length merely because an answer is detailed. If several assistants meet your conditions, compare their documented upload limits, current prices, data controls, and fit with your editing process. If none meets them, do not use the chat output for the final creative decision.
Where the plans differ
Chat Picker’s Claude vs ChatGPT vs Gemini matrix separates the figures into rows for Free plan, Low-cost tier, Main paid plan, Heavy-use plans, Team plan, Context window in the app, How usage limits are described, Training on your chats, and Ads.
In the October 2026 comparison, ChatGPT Plus is $20 per month, billed monthly. Claude Pro is $20 per month or $17 per month with annual billing, with $200 due up front. Google AI Pro is $19.99 per month. Confirm the vendor page before subscribing because plans and regional prices can change.
The context rows use different systems: Claude says context can reach 1M and varies by model, OpenAI lists figures by plan and model, and Gemini’s plans page gives no app token figure. The OpenAI and Claude pages give no exact message counts, while Gemini describes compute-based limits that refresh every 5 hours up to a weekly limit. These are documented plan differences, not measures of photography performance.