TikTok photo carousel comments take 9 photos: make the set in Sume

Yes: TikTok's September 3, 2026 announcement says people can add up to 9 photos in a single comment, with Photo Carousel Comments rolling out globally over the following month. A brand replying to a question can now answer with a short visual set, and Sume's Image API can make nine matching images in one planned batch.

The announcement is thin on specifications. It gives a count, not file sizes, formats or aspect ratios, so this post plans around what it states and flags the gaps.

What did TikTok announce for comments?

The newsroom post, titled "Say More in the Comments", introduces four things. Voice Comments let people record up to 60 seconds of audio under a video, rolling out globally over the next month to accounts 18 and older. Comment Polls let a creator add a poll to their own comment with up to 5 options and a voting timeframe, available globally. Photo Carousel Comments allow up to 9 photos in one comment, rolling out over the next month. Live Photo Comments add a short burst of motion to a photo comment and are available globally. TikTok adds that more than 1.7 billion comments were shared daily before these launched.

Rollout is staged, so the carousel option may not show on every account yet. Test on the account you will use, not a colleague's.

What does the announcement not tell you?

It gives no maximum file size, no accepted formats, no recommended aspect ratio, and no statement on how AI-generated photos in comments are labeled. Do not assume the photo-post rules for feed carousels apply here; those are a different surface with their own limits, covered in the linked posts.

Until TikTok publishes specs, choose conservative defaults: square images in a common format, kept small, and the same dimensions across all nine so the carousel does not jump between frames. If an upload is rejected, reduce size or switch format before anything else.

How do you build a matching nine-image set with Sume?

Use POST /v1/images. The catalog lists ChatGPT Image 2.5 as openai/gpt-image-2.5, which accepts up to 16 reference images, so you can pass your product shot or a style frame as an image_url entry in input_references and keep all nine on-model. The aspect_ratio parameter takes values such as 1:1 and 9:16, and output_format can be jpeg or png. Each model publishes the parameters it accepts on the models endpoint, so read that before you rely on a field.

Generate the set as separate calls with one changed prompt line per image, and keep a shared style sentence at the start of every prompt. Jobs can be asynchronous when a call runs past 30 seconds, so collect results through jobs and results.

curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "Studio photo, white background, soft light: the blue travel mug, lid open",
    "aspect_ratio": "1:1",
    "output_format": "jpeg",
    "input_references": [{"type": "image_url", "image_url": {"url": "https://media.sume.com/artifacts/artf_demo/mug.jpg"}}]
  }'

What should the nine images be, and what stays manual?

Treat the carousel as a mini answer. Image one answers the question, images two to eight show the detail, and image nine says what to do next. For a sizing question, that might be the product on a person, a measuring shot, three colour options, and a last frame with the product name.

Sume does not post comments. You download the images and add them in the TikTok app yourself, and you decide whether to say they are AI-generated in the comment text. The announcement does not describe a label for AI photos in comments, so state it in words if the images are not real photographs of the product.

Plan the batch before you generate. Write the nine prompts in a sheet, with the shared style sentence in the first column and the single changed line in the second. Run them in order and review each result against the one before it, so colour, lighting and framing stay consistent. If one image drifts, regenerate only that image with the same references instead of rerunning the set.

Keep the source files. If you reuse the set in a feed carousel later, the feed rules differ, and the linked photo-post articles list sizes and formats reported for that surface. Having the originals means you can resize for the new surface without a second round of generation.