AI Assistant
Chat that plans and runs AI Suite generations for you: it proposes first, you approve, then it generates.
AI Assistant (Beta) is a chat that turns plain requests into AI Suite generations. Paste your store link or attach your images, say what you want, and it plans the right tools and settings, shows them on a card, and only generates once you approve. Its strength is bulk: up to 500 products in one workflow on Studio plans (50 on Creator and free accounts), with on-model shots, new scenes and poses, ads, product videos and UGC clips for every product. Use it when you are not sure which tool fits or want a whole catalogue done. If you already know exactly what you need for a single image, the tool's own page works just as well.
At a glance
- You provide
- A request in chat, plus images, a video, a post link or your store link
- You get
- Inline images and videos in the chat, saved to your history
- Best for
- Not sure which tool to use, or reshooting many products, up to a whole catalogue
- Image formats
- JPEG, PNG, WebP only, up to 10 MB each
- Attachments
- Up to 6 images per turn; more become a bulk batch of up to 500 on Studio plans (50 on Creator and free)
- Videos
- MP4, MOV or WebM for the assistant to watch, subtitle or cut an edited video from: per message up to 3 videos of 3 minutes (free), 10 of 5 minutes (Creator) or 25 of 15 minutes (Studio)
- Product links
- Paste product page links; the assistant reads each page's photos, name and price
- Store links
- Paste your store or a collection; it brings in up to 100 products at a time as a numbered catalogue
- Status
- Beta
The assistant always proposes first. A generation is charged only when you click Generate on its card, or Start workflow on a workflow plan, and the button is blocked if your balance is too low. Chatting is free in Fast mode; Quality mode uses a few credits per message, which a one-time heads-up explains.
How it works
- 1
Open the assistant
Open AI Assistant from the sidebar, or start typing on the AI Suite home composer, which hands off into a chat.

- 2
Attach and describe
Add your image(s) with the + button, by pasting (Ctrl/Cmd+V), or by drag-and-drop, then say plainly what you want. For example: "edit this to a clean white background" or "put this dress on a model, then make a 5s video."
- 3
Review the proposal card
The assistant may ask a clarifying question, then docks a proposal card above the composer showing the tool, settings, input previews, estimated time, and credit cost.

- 4
Tweak and approve
Optionally adjust the card's knobs (model, quality, resolution, and so on); changing a knob reprices the card live. Click Generate to approve, or Cancel to dismiss.
- 5
See the result and iterate
The result appears inline; click it to open the full preview and download. Keep chatting to refine. Everything is saved to your Generations history.
- 6
Run a batch (optional)
Drop many images or paste your store link, say what every product needs ("for each of my products…"), review the Workflow ready to run card, choose Approve all or Step by step, then click Start workflow. The run produces a Workflow folder and a downloadable ZIP. See Workflows.
Bulk catalogue runs
This is where the assistant shines. Drop up to 500 product photos at once on Studio plans (50 on Creator and free accounts), or paste your store link, describe the whole catalogue pipeline in plain language, approve it once, and it produces the catalogue end to end while you step away.
In a single run you can chain steps like these across every product:
Turn each flat product into a polished on-model shot.
Move the best ones into a new scene, keeping the same model and outfit.
Generate several poses of each product (for example, 10 angles or looks per item).
Make an image ad for each product with your headline.
Produce a short product video for each item from its best shot.
Create a UGC-style video for each item for social and ads.
- 1
Drop your catalogue
Attach up to 500 images on Studio plans, or 50 on Creator and free accounts (JPEG, PNG, or WebP, each under 10 MB); anything over 6 images automatically becomes a bulk batch. Or paste your store or a collection link: the assistant brings the products in as a numbered catalogue, up to 100 at a time, and says how many came in and what it skipped.

- 2
Describe the full pipeline
Say what should happen to every item, for example: "for each product, put it on a model, make 10 poses, then a 5-second video and a UGC video." The assistant sees 4 sample photos and the true count, opens any other photo by its number when it needs to, and fans your instruction across the whole set.
- 3
Review the Workflow plan
The assistant docks a Workflow ready to run card listing every step, the per-item and total credit cost, and the estimated time.

- 4
Approve once and let it run
Choose Approve all to run the entire catalogue unattended, or Step by step to check each step's results before the next one starts, then click Start workflow.
- 5
Collect the finished catalogue
Outputs land in a Workflow folder under Generations, grouped per product, and Download all as ZIP saves the whole run. See Workflows.
- Run Step by step on the first product to lock the recipe, then re-run Approve all across the full set.
- Ask for per-product tailoring so poses, captions, or video scripts vary by item instead of being identical.
- Big runs spend real credits: the plan card shows the total before you approve, and anything that fails is refunded.
Settings & options
| Setting | Options | What it does |
|---|---|---|
| Response mode | Fast, QualityDefault: Fast | The pill next to Send. Fast is free and handles everything, including multi-step batch workflows. Quality uses the strongest reasoning for the trickiest requests and costs a few credits per message. Your choice sticks across chats. |
| Attach image | JPEG, PNG, WebP; up to 10 MB each; up to 6 direct, or a bulk batch of up to 500 on Studio plans (50 on Creator and free) | Add via +, paste, or drag-and-drop. 6 or fewer are normal attachments, good for multi-input single edits. More than 6 switches to a bulk catalogue batch: the assistant sees 4 samples and the true count, opens any photo by its number when it needs to, and fans your instruction across every image. |
| Attach video | MP4, MOV, WebM; per message up to 3 videos of 3 minutes and 100 MB each (free), 10 of 5 minutes and 300 MB (Creator), or 25 of 15 minutes and 1 GB (Studio) | Add via +, paste, or drag-and-drop, like an image. The assistant watches the video, sound included, when your request needs it: "what do you think of this ad?", "what does the voice say at the end?", "make one like this for my product". It also watches the videos it made for you when you ask about them. A video cannot be the input of an image edit or of a filmed video; attach a photo for that (subtitles and edited videos are the exceptions, see below). Watching is free in Fast mode and adds a little to the message fee in Quality mode. |
| Edited videos | Vertical 9:16; ten formats; up to 4 videos on one card; 480p or 720p for the presenter clip | Ask for a video with a person talking, or one made from your own footage: "a short video of someone talking about my cleaning service", "turn this clip of me into a proper video with captions", "make a video from these photos with a line on each", "a demo with a voiceover from this screen recording". The assistant edits it instead of filming a scene: your own photos, clips or screen recordings, typeset text and captions, a voice or music, and a closing card with how to reach you. Where a person is needed it films a short clip of a generated presenter, so a presenter talking to the camera needs no file at all. It works for any kind of business, with or without a website. Your files are read first and only what they really show is used. A file's own sound is not played, except your own talking clip, where your voice is the video. The first time a video needs a presenter the assistant asks who it should be (pick a look or describe one); the same person then presents every edited video of that brand, and is never shown as you, your staff or a customer. The card shows the format, every line, your files, the length and the cost before you click Generate; to change a line, tell the assistant. A video with no presenter costs 1 credit. With one it is 1 credit plus 0.5 credit per filmed second at 480p (1 per second at 720p, on paid plans), and a talking presenter is charged for the seconds really filmed, so the card's price is the most it can cost. It takes 1 to 2 minutes without a presenter and 3 to 8 with one. The formats are listed in Edited video ads. Finished videos appear in the chat and in Ad Studio under My Ads. |
| Carousels, your screen on a device, your photo framed | A carousel of 3 to 5 slides (3 on free accounts); 4:5 or 1:1 for a carousel, also 9:16 for the other two | Three kinds of image that go beyond a single edit. A carousel: ask for one ("a 4 slide carousel for my cleaning business, set in this photo of my kitchen work") and it is made as one wide picture cut into the slides, so each swipe continues the last, from your own product photos or from your description, with your exact words on each slide. It costs 1 credit a slide, and a paid plan's free images pay first. Your screen on a device: ask to show your website or app on a phone or laptop ("my site on a phone held by a woman in a cafe") and give the page address or attach a screenshot. Your real page is placed on the device exactly as it is, so its words are never redrawn. It costs 1 credit, or one of your free images. Your photo, framed: ask to use a photo as it is for a placement ("use this photo unchanged for a story"). Nothing is generated or changed: the photo is cropped around its subject when that fits, or shown whole, and it costs nothing. Each one shows a card first, and to change it you tell the assistant. A single image ad or an edit of a photo is still a Smart Edit. Finished images appear in the chat and in Ad Studio under My Ads. A carousel or a screen image also adds an entry for the product or the business it was made for in Ad Studio under Products. |
| Subtitles | Several videos per card (by plan); nine caption looks or your own; headlines and key-word text; any language; a .srt file | Ask for subtitles on a video you attached, a post link, or a video the assistant made: "add subtitles to this", "captions on all of these", "Spanish subtitles". The spoken words are transcribed and placed where they cover no face, product or text. The card shows a frame with the look and every caption line: edit any line before you click Generate. To change them later, just say what to change ("smaller", "no yellow", "fix the brand name"); the new version is made from your original video, never on top of the old subtitles. Not sure which look you want? Say so and the assistant shows the styles to pick from, or attach a video whose text you like and it copies that look. Besides captions it can add other text: a headline, key words that pop up as they are said, a closing line with your link ("add a headline Summer Sale for the first 3 seconds", "text like this video"). If one caption or line sits in the wrong place, say where ("at 0:05 it covers her face"): the assistant looks at the video and moves just that one. New subtitles on a video cost 0.5 credits up to a minute, plus 0.25 for each further minute (free accounts pay double); changing them later is free, and so are transcripts on their own (Quality mode's message fee still applies). Videos without speech have no captions but can still get headlines and other text. |
| Video post links | Instagram, TikTok or Facebook post or reel; videos up to 5 minutes | Paste the link instead of downloading the video: "what makes this reel work?", "make one like this for my product". The assistant fetches the video once and keeps it in the chat, so you can ask about it again later. YouTube links cannot be fetched; download the video and attach it instead. |
| Product page links | Any public product page; several links in one message | Paste the links and say what you want: "review the photos on my product page", "make ads for these 4 products", "one Seedance reel with all of these". The assistant reads each page's photos, name and price, picks the right photo per product and tells you which one it used. Asked for a review, it says when photos are already good and suggests changes only where it can point to the problem. Reading pages is free; the generations are charged as usual, after you approve the card. If a store blocks reading or the link is broken, it says so and asks for the photos. |
| Your store or a collection | A store or collection link; up to 100 products at a time, and up to 500 photos in one catalogue on Studio plans (50 on Creator and free) | Paste your store or a collection link and say what you want: "a modeled photo of every product in this collection", "redo my store's photos in three poses, with a video of each". The assistant reads Shopify and WooCommerce product feeds, collection pages and sitemaps, brings each product's photo into the chat as a numbered catalogue named after the product (for new shots it picks the photo with the plainest background), and plans one run over all of them. It tells you how many products came in and which it skipped and why, and a product already in the chat is not brought in twice. To only see what a store sells, ask "what's in this collection?": nothing is copied. Ask "are my store's photos any good?" and it reviews about 20 products (or the ones you name), says which photos to keep and gives a reason for every change. Generations are charged as usual, after you approve the card. |
| Pinned references | Model, Background, Pose; up to 8 pinned images | Click Reference next to + in the composer and pin a photo as the Model (the person to show in every result), the Background (the setting every result is placed in) or a Pose (the pose or framing to match, one output per pose). A pinned reference is used with every product in the chat until you remove it, so a whole catalogue keeps the same model or setting. Model and Background hold one photo each; Pose can hold several. |
| Web research | Searches the web and reads public pages | Ask the assistant to look something up before it plans: "research what catalogue photos work for my category, then plan the shoot", "what are this marketplace's main image rules?". It searches, reads the pages it needs, and gives you the findings with their links, then says which finding shaped which choice in the plan. It searches only when your request needs it, and each account has a daily research limit. |
| SVG files | A logo, icon or simple illustration as an .svg | Ask for an SVG ("make an SVG logo of a fox in two colours") and the assistant draws one with GPT Image 2.6 from your description, at the same credit cost as the same picture as a PNG. The result shows a PNG preview with Download SVG under it, and SVG is offered among the preview's download formats. An SVG is drawn from words, so a photo or an existing logo cannot be converted or traced into one. |
| Proposal card knobs | Model, quality, resolution, aspect ratio, duration, and so on (varies by tool) | Editable on the docked proposal card before you approve. Only legal options for the chosen model are offered, and changing a knob reprices the card live. |
| Workflow run mode | Approve all, Step by step | Offered only for multi-step plans, on the Workflow ready to run card. Approve all runs the whole batch on its own. Step by step runs step 1, then waits for your OK before each next step. Nothing starts until you click Start workflow. |
Costs show live in the credits this session pill (hover for the split):
- Chat (text): free in Fast mode; a small fee per message in Quality mode.
- Generations: the real per-image or per-video cost, deducted only when you approve a proposal. The exact number is on the card.
- Prompt tailoring: for bulk runs with per-product prompts. Free in Fast mode; in Quality mode an approved ceiling billed at actual usage.
Failed generations are refunded. Workflow steps are charged only as they run; steps that never run are not charged.
Best practices
Attach the actual image and say plainly what you want; let the assistant propose settings, then adjust the card's knobs rather than over-specifying in text.
Start in Fast (free) for everything, batches included; switch to Quality when a request needs the strongest reasoning.
For catalogues, phrase it as "for each of my images…" and drop the whole set so one instruction fans across every product.
Review the proposal card's cost and estimated time before approving, especially for big batches.
Pick Step by step for a new batch recipe so you can check step 1's output before spending on the rest, then switch to Approve all once you trust the recipe.
Keep unrelated tasks in separate chats to keep context, and cost, down.
Use Revert to here to branch from an earlier prompt instead of retyping.
Troubleshooting
Only JPEG, PNG, and WebP are accepted. Convert the photo first, then attach it.
Each file is capped at 10 MB. Resize or re-export the image smaller and try again.
Videos must be MP4, MOV or WebM and within your plan's limits: up to 3 videos of 3 minutes and 100 MB each per message on the free plan, 10 of 5 minutes and 300 MB on Creator, 25 of 15 minutes and 1 GB on Studio. Hover the + button to see yours. Trim or re-export the clip, then attach it again.
A proposal is waiting for your decision, or a generation is running. Approve or cancel the docked card, or wait for the result, to continue.
That is expected for bulk catalogue runs. The assistant sees 4 samples and the true count, opens any other photo by its number when it needs to, then applies your instruction to all of them when the workflow runs.
A catalogue holds up to 500 photos on Studio plans and 50 on Creator and free accounts, and the assistant says what it skipped and why: out of stock, already in the chat, or no usable photo (missing, too small or a banner). For a big store, ask for one collection at a time, or name the products you want.
The proposal costs more than your balance. Top up credits, or edit the card to a cheaper model or quality, then approve.
Uploaded images are reclaimed after a while. Re-upload the image(s) and ask again.
Long chats carry more context per message. Start a New chat for unrelated work; it is cheaper.
Hover your message and click Revert to here. It removes that message and everything after it. Your already-generated images and videos stay in your gallery.
What to do next
Keep going with what the assistant makes: