AI Product Video From a Text Prompt: No Listing URL, No SKU Photo Required

AI product video from a text prompt used to mean "type a description, pray the tool understands ecommerce." Most AI video tools require a URL or an uploaded photo before they let you render anything, because their pipeline is built around a reference frame. AmazVid just opened the third door: no URL, no photo, just your words — routed through the same MiniMax H3 video pipeline and image-01 image pipeline that the URL and photo paths use.

This is not a lightweight demo mode. Prompt-only requests get the full render treatment, deduct the mode's normal credit cost, and land in your Library exactly like a URL-based render. If you are an agency pitching concepts, a dropshipper testing hooks before shipping inventory, or a DTC brand mood-boarding a seasonal campaign, this changes the workflow.

Start here: Open the launcher · Prompt tab in the wizard · Pricing · AmazVid AI product video generator.

Why prompt-only exists

The tacit assumption in every AI product video tool from 2024–2025 is: sellers already have the product photographed. If you have inventory, that assumption holds. But three groups had no path in until now:

• Creative agencies pitching a new client — they have no SKU photos yet; they need to visualise a concept before the client signs.
• Dropshippers and pre-launch brands — they want to test ad hooks and thumbnails on paid social before ordering inventory that might not sell.
• DTC merchandising teams planning seasonal drops — they want mood boards, style tests, and internal-review clips for Q4 or spring release weeks before product photography is scheduled.

For all three, the "must have a photo" gate was a hard block. Prompt-only removes it. The credit cost is unchanged because the render cost is unchanged — a MiniMax H3 job takes the same GPU seconds whether the input is a photo or text.

What actually happens on the backend

When you type a prompt and hit Generate, three things run:

1. API schema (`validators/video.ts` and `validators/image.ts`) accepts an item with just a title (no URL, no image URL). The old refine rule was "URL or (image + title)" — now it also accepts "title alone."
2. API route (`api/videos/generate/route.ts`) skips the SSRF check on the upload prefix when there is no image URL at all, marks source_type as 'manual', and enqueues the job.
3. Render pipeline (`inngest/functions/process-video.ts`) detects the missing image and skips the first-frame preparation entirely. It hands MiniMax H3 a content array with just a text part — the model interprets that as text-to-video and generates the shot from your prompt alone.

The same code path handles all four modes. For Ad Image, the image pipeline (`inngest/functions/process-image.ts`) skips the product-chip composite when there is no product image buffer, and generates the background straight from your text via MiniMax image-01 intl.

The two entry points

You can send a prompt-only request from two places on AmazVid:

1. The shortcut launcher — fastest path

The `/dashboard/new` page has a purple gradient card sitting between the main wizard and the Template Gallery. Headline: *"No product photo? Generate from a prompt."* It shows all four modes as pills with credit costs printed on each tab. Pick a mode, type your prompt in the textarea, hit Generate. That is one screen, no wizard steps.

Under the hood, the launcher calls the same API endpoints the wizard uses. The result: your credits deduct exactly the same way, your job lands in your Library exactly the same way, and you can Download / Try again in the exact same UI.

2. The Wizard's Prompt tab — full control

If you want to layer camera motion, lighting, background, and a Director Brief on top of your text, use the wizard's Step 1. There is now a third tab next to Link and Photo: Prompt. Selecting it swaps the URL input for a large textarea. Continue to Step 2 to pick the cinematography knobs. Step 3 confirms cost. Step 4 renders.

The wizard route is slower (four steps vs one) but gives you every parameter. Use it when you have a specific shot in mind.

Prompt writing that gets a usable clip

Prompt-only quality is 80% about the prompt. After running hundreds of test renders, the pattern that works:

Subject. Who or what is the frame anchor? Named age, gender, ethnicity, garment or product all help. "A 30-year-old woman in a soft grey hoodie" beats "a person" every time.

Setting. Where is it? "Warm morning kitchen with marble counter" beats "a room." Specific props anchor the model to a coherent scene.

Camera. What is the camera doing? "Slow push-in from the side," "low-angle tracking," "overhead flatlay descent." Verbs matter.

Light and grade. "Golden hour warm cinematic," "cool neon-lit night," "soft north-window flatlay." One phrase decides the whole mood.

Aspect. Prefix the prompt with "10s vertical 9:16" or "square 1:1 static." MiniMax H3 respects framing hints when they lead the prompt.

Sample that works:

*10s vertical 9:16 UGC-style Meta hook. A 28-year-old man with a five-o'clock shadow sits in his driver seat in warm afternoon sunlight, seatbelt on, casual grey hoodie. He holds a protein snack bar in kraft wrapper up toward the camera in a selfie framing, blurred dashboard and windshield behind. Slight handheld wobble, iPhone-look color grade, honest customer-review feel.*

That is 65 words. Ad Creative mode, 10 seconds, 5 credits. The render lands a Meta-native UGC hook you can iterate on before you have the actual product to hold.

The credit-cost table

Prompt-only bills the same as URL and photo inputs. Nothing is discounted or surcharged.

The free plan includes 8 credits — enough to test one Ad Creative concept, one 5s plus one 10s Product Shot, or a few Ad Images at 2 credits each. Free-plan renders carry a small watermark; a one-shot clean MP4 export is $3. Paid plans start at $29 / month.

Three workflows that this unlocks

Workflow 1: Agency pitch in an afternoon

Client brief lands at 10am. By 2pm you have four concept videos:

• 11:00 — Prompt 1 (Ad Creative, 5 credits): a UGC unbox POV
• 11:30 — Prompt 2 (Ad Image, 1 credit): a Meta feed hero still
• 12:00 — Prompt 3 (Product Shot 10s, 4 credits): a slow luxury orbit
• 12:30 — Prompt 4 (Wearable, 6 credits): a lifestyle try-on

Total: 16 credits (~$4 of pool). Total time: ~4 hours of iteration. Total photo shoots required: zero. When the client picks a direction, upload the real SKU photo and re-render at full fidelity.

Workflow 2: Dropshipper concept-testing before ordering

You are considering three products for Q4. Instead of ordering samples of all three and running paid social tests on the winning one, you run prompt-only Meta hooks for each and A/B them on a small test budget. The one that scrolls-stops best gets the actual inventory order. Reduces sunk cost on unsuccessful SKUs.

Workflow 3: Seasonal mood board for a DTC brand

Your merchandising team is planning spring 2027 drops in October 2026. Product photography is scheduled for January. In October, generate 20 Ad Images (20 credits, ~$5 pool) with seasonal scene prompts — "spring cherry blossom flatlay," "summer beach product still," "autumn cozy latte desk." Circulate to marketing, review, iterate, lock the aesthetic. When January photos arrive, re-render with real SKUs on the pre-approved look.

What prompt-only does badly

Prompt-only invents the subject. That means:

• Product identity is not locked. MiniMax generates a plausible product that fits the prompt, not YOUR product. If you write "black wireless headphones with silver accents," you get plausibly-black-plausibly-silver headphones — colour, logo, band shape all invented.
• Wearable loses its main superpower. The whole point of Wearable is Ref2VA character-lock — your actual garment on a model. Prompt-only Wearable is "guess at a garment on a guess at a model."
• Amazon listing videos need honesty. Amazon's Seller Central requires the video to match the actual product. Prompt-only is fine for ads and content, unsafe for listing gallery media.

When to graduate from prompt-only

Prompt-only is a testing gear. Real production creative should upload the SKU photo. The rule of thumb:

• Prompt-only: concept testing, agency pitches, mood boards, pre-launch tease
• Photo + prompt (Director Brief): production ads, seasonal campaigns, iteration on a real SKU
• URL only: batch catalogue runs where the listing image is authoritative
• URL + Wearable: fashion, footwear, eyewear that need on-body Ref2VA lock

Most brands use all four over the lifecycle of a product.

Comparison to other tools

The distinction: AmazVid wraps prompt-only in an ecommerce-native flow (mode → credit → Library → Download → re-render), so a prompt render is directly comparable to a URL render on the same product.

Next steps

If you have a listing already, keep using URL or Photo — that gives you the best product fidelity. If you are pre-launch, agency, or dropshipping: try prompt-only for concept testing.

Try it: Prompt-only launcher · Pricing · Ad Creative from photos · Ad images from listing photos · Cost savings vs studio.