Creative Space
Turn Saved X Demos into Creative Experiments

Contenuto

Provide: Your X account (I will read your bookmarks and timeline after authorization), or the account and keywords you want to follow, or several links to posts you want to try.

Receive: The recurrence conclusion of each demo, word prompt cards (original and rewritten), comparison results of trial runs with Flova, entries saved in the inspiration library; optional: regular candidate list.

X has an official MCP, use it for reading first, there is no need to choose between MCP and computer use. However, the connection may require your own X developer application, and some hosts may not be able to connect. See "Connecting to X" below. For browser-only platforms such as Xiaohongshu and Douyin, use "Social Media Inspiration Studio".

Three entrances

  1. My bookmarks and timeline: Read the bookmarks of the last 7-14 days, organize them into "recent attention", and select the ones you can try from them. Only look at the most recent paragraph to serve current interests and do not make long-term portraits.
  2. Search, trend, follow accounts: Find new demos by keywords, accounts (competing products or interested people), and regional trends, and make a candidate list.
  3. Single post: Accept, read clearly, test run, and store.

Tools and responsibilities

  • X Official MCP: Read posts, searches, user timelines, bookmarks, trends. Recommended, see "Connect X" below.
  • Teardown — retrieve and read the full “Break Down a Reference Video into an Interactive Report” inspiration through Flova MCP by name: Disassemble the structure, vision and movement of the actual work on demand; the text of the post and external links are obtained by the reading tool available to the host.
  • Flova: Query model capabilities and quickly generate original and rewritten versions for trial runs. Required.
  • inspiration library: User's Notion, Feishu multidimensional tables, or local files. Required.

Method index

  • : How to read each of the three entry routes, what to extract from posts, demo types, and reproduction difficulty. Read in steps 2–3.
  • [Read demo from Steps 2–3 Read.
  • : card format, reference assets upload and role, trial generation comparison. Steps 4–5 Read.

Connect X

The official MCP address is https://api.x.com/mcp. The configuration method and tools are subject to the official document: https://docs.x.com/tools/mcp. A few things that are easy to step on:

  • **Reading bookmarks and personal timeline require user identity. ** When filling in only the application-level Bearer token, you can only read public posts, searches, and trends, but not bookmarks. To bring user identity, you need to use the official xurl local bridge, which requires the user's own X developer application.
  • Host authorization support: Before connecting a user-identity version, check that the host supports the required OAuth authorization. If it only accepts an MCP URL, first test whether authorization can complete.
  • When the connection is unavailable: Users can post links or screenshots, or use "Social Media Inspiration Studio" to read them in a browser. No developer applications are registered for users and no fees are paid.

Fees

  • Flova: The trial run consumes the user's quota, please explain first.
  • X: Developer applications may need to be upgraded to pay-as-you-go (required for connecting to Times client-not-enrolled). The fee is paid to X and has nothing to do with Flova; explain it first and then configure it.

Boundaries

  • Read only: No posts, no likes, no reposts, no followings, and no bookmarks or deletions for users.
  • Prompts are the author's work: Note the author and post link; paywalls, private messages, and content that requires attention are not bypassed and are decided by the user.
  • Do not reprint: The inspiration library only stores links, prompts and the user's own test results; other people's pictures and videos are only compared on this machine.
  • Results not guaranteed: Posts often show the best results after selection, and may also go through later stages. Do not write "what the author said" as "verified".
  • Token is secret: Do not read, do not display ~/.xurl and token.

Workflow

**Confirm the connection (confirmation point) → Read parallel candidates (confirmation point) → Identify and determine whether it can be reproduced (confirmation point) → Change to user (confirmation point) → Test run comparison → Store → Regular tracking (optional) **

1. Confirm connection (confirmation point)

Follow “Connect X” above: inspect available tools and determine whether X MCP uses app-only Bearer or user authentication. If absent, explain setup steps/fees or use post links/screenshots. Bearer cannot read bookmarks; state that limitation and use route 2 or 3.

2. Read parallel candidates (confirmation points)

  • Entrance 1: Bookmark, organize 3-5 "recently followed" topics, and provide users with confirmation.
  • Entrance 2: Read according to keywords, account numbers, and trends. After screening, 5-10 candidates will be given. For each entry, write the link, account number, one sentence, type, difficulty, and whether there is a prompt.
  • For details, see . The user selects what they want to try from the list.

3. Identify and determine whether it can be reproduced (confirmation point)

  • Organize each item: author, link, author's tools and models, original prompt text, parameters, pre-assets, works, and type.
  • Use creation_generation_models (detail="full") to check the duration, aspect ratio, number of reference assets, and roles of the corresponding model. You cannot find the word "Unverified".
  • There are four conclusions: try directly, try after rewriting, need other tools, cannot be reproduced (write down the reasons). Only when the work does not have a prompt, use the method of "link copy prompt" to remove it.

4. Change to user (confirmation point)

Keep a copy of the original text of the prompt as a comparison benchmark; write another version and rewrite it: replace the subject with the user's assets, adjust the writing method according to Flova's existing model, and remove the author's brand, real name, and signature. Use format to indicate the original text or rewriting.

5. Test run comparison

After the user agrees: run the original version first (if it can be run as is), then run the rewritten version; the results will be shown to the user side by side with the original work, explaining where they are similar and where they are different. Only one variable is changed in each round, and after trying two more rounds at most, it still cannot be recorded as "cannot be reproduced".

6. Storage

Store it in the user's inspiration library: Notion, Feishu multidimensional table or local inspirations.md. The user will be asked where to put it for the first time, and will not be asked again after that. One line at a time: name, post link and author, type, author's tools and models, original text of the prompt (reverse push is marked "reverse push"), rewritten version, required assets, test run results and comparison conclusion, experience, tags and status. Only links, prompts, parameters and the user's own test running results are stored. Other people's pictures and videos are only stored on this machine. Check for duplication before entering the database: the same link will not be added when updated, and the same prompt will be forwarded by multiple people and merged into one.

7. Regular tracking (optional)

When the user wants to follow an account or keyword, confirm the account, keyword, focus, frequency (daily or weekly), maximum number of entries at a time (5–10 is recommended) and where to send the list, and save it on the "Tracking Settings" page of the inspiration library (or tracking.md). After the setting is confirmed, the guide is set as a scheduled task:

  • Verify X MCP access in the scheduled environment first: an application Bearer token can run unattended but cannot read bookmarks; the xurl local bridge requires the computer to remain on.
  • Run it manually first, and then set it if appropriate after the user has reviewed the list.
  • The task content is self-contained (start from scratch every time, do not remember the dialogue): read tracking settings, read X's new developments according to settings, filter and send the candidate list, and update the last run time.
  • Use the scheduled function that comes with the host to create it (whether each host supports it and what it is called depends on the actual situation); if there is no creation tool, write the task content and let the user create it by himself.
  • First confirm that the MCP of
  • Only the candidate list will be given each time, no trial run, no storage, and no consumption of Flova credits.

Delivery

Recognition results and recurrence conclusions, word prompt cards, test run comparisons, inspiration library entry links are attached; if you have used Entry 1, "Recent Follows" is attached; if you have followed up, settings and the first batch of candidate lists are attached.

Project directory

x-demo-<日期>-<名称>/
├── source/   帖子截图和视频(仅本机对照)、用户素材
├── docs/     近期关注、识别结果、提示词卡
├── trial/    试跑结果、对照图
└── library/  灵感库条目(未使用 Notion 或飞书时)

Read demo from X

Entry 1: Bookmarks and Timeline

  • Read the last 7–14 days, up to 30–50 bookmarks, user can adjust. Bookmarking usually means "want to try it myself". Read the timeline of the watchlist again, the window is the same. Don’t read earlier, don’t read private messages.
  • Each note: link, author, sentence, content type (tutorial, demo, case, inspiration picture, other).
  • Summarize 3-5 topics, each with 2-3 examples collected by the user; mark which ones you want to do yourself (with reminder words, reproducible). Give users confirmation and modification.
  • Only the topic list and post links are retained, with dates, and will be overwritten by the next refresh. Text and media are not saved.

Entry 2: Search, trend, follow account

  • Search: Use keywords to add "prompts" features (prompt, prompts, workflow), and then limit the media according to the form of the work.
  • Account: For competing products or interested people given by users, read their new posts since the last time. For competing products, write an extra sentence or two about dynamic observations: what topics were posted, which posts had higher interactions, and whether the format has changed. Only write what you observed.
  • Trends and News: Read by region, only as a clue for finding directions, not as a demo.
  • Filtering: Those with prompts or clear workflows are ranked first; only finished products are downgraded without any method; obvious advertisements, traffic diversions, and paid traffic diversions are excluded; those already in the inspiration library are excluded; those whose authenticity cannot be seen are marked "unverified".
  • Each item in the candidate list: link, account, one sentence, type, difficulty of reproduction, whether with prompts, and why it is recommended.

Entry 3: Single post

What to take out:

ItemHow to find
Author and linkUsername, post link, publishing time
PromptsMain text; the author's top reply and several subsequent items (read the entire series of posts, and the prompts are often at the back); the text in the picture is transcribed verbatim and allowed to be checked by the user; external links (Gist, Notion)
Tools and modelsThe author’s name, hashtags, and watermarks on the screen
Parametersaspect ratio, duration, resolution, seed, remember only after writing
Front-end assetsReference pictures, reference videos, and audios, have they been provided by the author
WorksPictures and videos; video uses Teardown to obtain the content and then extract a few frames to see the key scenes
CommentsResults reproduced by others, problems encountered, and the author's additions

If it cannot be found, write "Post not provided" and do not speculate.

demo type: ① Prompts are given directly; ② Prompts plus reference assets (the author of assets does not provide them, please use your own instead); ③ Multi-step workflow (split into steps, corresponding to the available tools); ④ Only works (use the method of linking to copy the prompts, the result is marked as "reverse push"); ⑤ Display type (note down for reference, no test run); ⑥ Paid or offered privately (source documented, not bypassed).

Reproduction Difficulty: Easy (single generation, complete prompts, the author's model Flova has or is similar); medium (multiple reference pictures, specific parameters, and abilities are not completely corresponding, rewrite and try); difficult (multiple tools are connected in series, post-processing or 3D is required, and only the parts that can be tried are tried); Impossible (no corresponding capabilities, assets are not available, the results are obviously subject to a large number of selections or post-processing, and are recorded truthfully).

Judge the credibility of the work: Is it showing a single result or a selection? Does the prompt match the work? Is there anyone in the comments who has successfully reproduced or failed? Does the video have obvious post-production (subtitles, transitions, color correction), which are not caused by the prompt.

Prompt Cards and Flova Handoff

These inspirations deliver prompt cards, not images/videos: one per image or shot. Cards can be used directly for generation; this reference covers writing, Quick generation, and validation.

Card format

Include all fields; write “none” where absent:

FieldContent
IDE.g. P01, mapped to the requested image/source shot.
Model/capabilityE.g. Seedance 2.5 multimodal video; GPT Image 2 editing.
PurposeOne-sentence intended use.
PromptComplete copy-ready text in one code block.
ReferencesOne row each: filename, submission order, role below, responsibilities, exclusions.
ParametersDuration, aspect ratio, resolution, etc., from creation_generation_models; otherwise “unverified.”
InvariantsUnchanged people, products, text, composition.
RationaleTwo/three sentences linking choices to model-writing rules.
Acceptance checksE.g. identity, verbatim text, absence of teleporting motion.

+Write in the target language. Chinese-user requests default to Chinese unless the model guide/user calls for English. Preserve dialogue/on-screen text verbatim; never translate/polish it.

Quick generation

+Follow Flova Skill's Quick/reference guidance:

  1. Query the model through creation_generation_models, detail="full"; verify parameters, reference slots/roles.
  2. Prepare references: supported aspect ratio/padding, crops, formats.
  3. creation_upload_ticket → transfer bytes to its URL → creation_upload_complete → file_ref. It expires in roughly one hour; upload again before submission when necessary.
  4. Assign each inputs role: first_frame starts matching that image; last_frame ends on it; reference guides appearance/style/content without exact matching; driving_audio drives lips/action; base_video is edited/continued.
  5. Submit with client_request_id; retain card/reference/file_ref/task-ID mappings.

Referencing inputs in prompts

  • Use ordinal natural language, e.g. first image controls face/clothing, second controls product. Match inputs order within each media type; images/videos/audio each count from first.

Other tools

+Remove Flova-specific markers/role instructions and use the destination tool's upload mechanism and parameter controls. Explain that these prompts target a particular model; changing models may change outcomes.

Optional comparison

+Explain quota use before proceeding. Generate original and revised prompts once each with identical parameters/references; show side by side. Change one variable per round (e.g. reference roles or timeline), record edits/task IDs/results, and deliver the record. Generation is stochastic: one comparison shows a tendency; repeat important findings once or twice before concluding.

Processo creativo

  1. X

    X

    Find creative approaches

    Read posts from bookmarks, selected accounts, or keyword searches.

  2. Flova

    Flova

    Create your own version

    Try the selected approach with your own images or videos.

  3. Notion

    Notion

    Save the inspiration

    Record the original post, prompts, and actual results.