Qstar Video Ecom

Use this skill when the user wants to make a Chinese e-commerce short video quickly, especially for Douyin, Xiaohongshu, Taobao, Pinduoduo, JD, or WeChat Video.

Default value: keep the old one-click product video flow as the stable path, then layer newer video-factory style modes on top.

What this skill supports

This skill now has 4 modes:

1. 1. INLINECODE0

Classic one-click product video. Best when the user says "做个产品视频", "生成电商视频", "帮我出一条带解说的视频".

1. 2. INLINECODE1

Single-image fast start. Best when the user already has one product image, poster, hero shot, or reference visual and wants to start from it.

1. 3. INLINECODE2

Batch multiple short-video variants from one product brief. Best when the user wants several hooks, tones, or platform versions in one go.

1. 4. INLINECODE3

Natural-language revision mode. Best when the user already has a generated video and says things like "把背景换掉", "节奏快一点", "换成男声", "再来 3 个版本".

If the user does not specify a mode, use classic.

Important boundary

Do not pretend the backend already has magical edit APIs if it does not.

For newer modes:

• - If the current bot.guanma.top/sora-api endpoints can support the request with the existing generate flow, use them.
• If the request exceeds current API capability, convert the user intent into a structured generation brief and explain that you are using the nearest supported workflow.
• Prefer a truthful fallback over a fake "done".

Step 0 — Extract USER_ID

Extract platform and sender_id from the current session key and convert to {platform}:{sender_id}.

Examples:

• - agent:main:telegram:direct:5239705501 -> INLINECODE8
• INLINECODE9 -> INLINECODE10
• INLINECODE11 -> INLINECODE12
• INLINECODE13 -> INLINECODE14
• INLINECODE15 -> INLINECODE16

Always inline the real value into commands. Do not rely on shell variables.

Step 1 — Detect mode

Route by user intent:

• - INLINECODE17

User wants one finished e-commerce short video and has no advanced request.

• - INLINECODE18

User gives or mentions a specific product image / hero image / visual reference and wants to start from that one image.

• - INLINECODE19

User asks for multiple versions, multiple hooks, multiple platforms, multiple tones, or "批量跑素材".

• - INLINECODE20

User already has an existing video and wants changes in plain language.

If mixed, prefer:
revise > batch-variants > single-image > INLINECODE24

Step 2 — Check profile

CODEBLOCK0

If onboarded: false, run the onboarding flow before generation.

Opening message:

CODEBLOCK1

Ask:

1. 1. INLINECODE26
2. INLINECODE27
3. INLINECODE28

Then:

CODEBLOCK2

Step 3 — Mode-specific intake

Mode A: classic

Follow the stable old flow:

1. 1. Ask target platform
2. Ask script style template
3. Ask product name + 2-3 core selling points
4. Ask voice choice
5. Check quota
6. If paid, allow product image upload
7. Generate one finished video

Mode B: single-image

Ask for:

1. 1. Target platform
2. Product name
3. One-sentence video goal
4. 2-3 selling points
5. Voice choice
6. One image upload or image URL

Then convert the brief into a structured product field like:

INLINECODE32

If the image is available and the user has paid quota, prefer passing a direct public image_url.
Only use the documented upload endpoint if it is confirmed live in the current environment.
If image upload is unavailable, say you are falling back to the standard image-assisted generation path.

Mode C: batch-variants

Ask for:

1. 1. Target platform or platforms
2. Product name
3. Core selling points
4. Number of variants
5. Preferred dimensions of variation:

- hook - tone - audience - CTA - voice

Then produce a variant plan first, for example:

• - Variant 1: price-shock
• Variant 2: bestie recommendation
• Variant 3: curiosity bait

Run the current generation path once per variant if the backend only supports single-job generation.
Do not claim server-side batch execution unless it truly exists.

Mode D: revise

Ask for:

1. 1. Existing video link or generation context
2. What to change, in plain language
3. Whether the user wants:

- overwrite-style replacement - one more revised version - several revised versions

Translate the user's request into a structured revision brief:

• - scene/background change
• faster/slower rhythm
• stronger CTA
• female/male voice
• platform-specific format
• more dramatic / more everyday / more premium tone

If no true edit API exists, tell the user you are generating a revised new version instead of frame-accurate editing.

Step 4 — Platform choice

Use this mapping:

CODEBLOCK3

Prompt:

CODEBLOCK4

Step 5 — Script style template

For classic and most batch-variants cases, use the proven template system.
Template details live in TEMPLATES.md.

Prompt:

CODEBLOCK5

Map:

• - INLINECODE38
• INLINECODE39
• INLINECODE40
• INLINECODE41
• INLINECODE42
• INLINECODE43

When the user chooses 1-5, auto-build the product brief using the template style plus platform.

Step 6 — Voice choice

Prompt:

CODEBLOCK6

Map:

• - INLINECODE44
• INLINECODE45

Step 7 — Quota and image upload

Check quota:

CODEBLOCK7

If credits > 0, allow image upload:

CODEBLOCK8

If the user sends an image, first obtain a public image URL through the host platform's image handling path, then:

CODEBLOCK9

Read image_url from the response.

If that endpoint returns 404 or equivalent unavailability, do not loop on the upload flow. Pass the original public image URL directly in the generate request instead.

The image URL must be reachable by the generation backend, not just by the local machine. Avoid internal-only hosts or asset domains that fail public DNS resolution.

If free user or skipped upload, keep image_url empty.

Step 8 — Generate

Tell the user generation is running:

CODEBLOCK10

Then:

CODEBLOCK11

Step 9 — Response handling

If success:

• - With uploaded image:

INLINECODE50

• - Free quota no image:

INLINECODE51

• - Paid quota no image:

INLINECODE52

For batch-variants, summarize all successful outputs clearly by variant label.
For revise, explicitly say this is a revised version if you had to regenerate rather than edit in place.

If HTTP 402 or quota exceeded:
run the payment flow.

Step 10 — Payment flow

Show:

CODEBLOCK12

Then ask payment channel:

CODEBLOCK13

Create order:

CODEBLOCK14

Read order_id, amount, qr_url.
Send the QR image first, then:

CODEBLOCK15

Poll every 10 seconds, up to 18 times:

CODEBLOCK16

If status == "paid":

CODEBLOCK17

Then return to Step 7.

If timeout:

CODEBLOCK18

Operating rules

• - Prefer the stable classic mode unless the user clearly wants a newer mode.
• Use single-image when the image is the center of the request, not just an optional add-on.
• Use batch-variants when the user cares about scale or testing multiple hooks.
• Use revise when the user wants changes to an existing output.
• Be explicit when a "revision" is actually a new regenerated version.
• Never hide quota/payment boundaries.

Data and privacy

This skill is provided by Q-star via https://guanma.top.

• - Transmitted data:

- product description text - optional public product image URL - platform-linked user ID - optional contact info

• - Transport:

- HTTPS API calls to bot.guanma.top

• - Storage:

- generated videos are stored in Qiniu for delivery

• - This skill does not read local files directly
• This skill does not access system credentials
• Payment QR codes are generated by the payment provider bound to Q-star services