使用 Fal 规划并生成风格统一、忠于来源的移动端信息图系列。
Article Infographic Series Turn an article or structured brief into a coherent series of mobile-first infographic images, review the page plan, and generate each approved page through Fal. Loadout capabilities Use direct:faltools for image generation. The primary operation is direct:faltools:faltexttoimage, which currently generates one 1K Nano-banana 2 image per call and saves it to the asset library. Inspect the live schema and price before execution.
# Article Infographic Series Turn an article or structured brief into a coherent series of mobile-first infographic images, review the page plan, and generate each approved page through Fal. ## Loadout capabilities Use <integration-tag>direct:fal_tools</integration-tag> for image generation. The primary operation is `direct:fal_tools:fal_text_to_image`, which currently generates one 1K Nano-banana 2 image per call and saves it to the asset library. Inspect the live schema and price before execution. The current catalog description lists $0.08, or 8 Aident credits, per image. Runtime billing and confirmation remain authoritative. ## Inputs Collect: - Source article or brief. - Audience and distribution surface. - Desired image count, or permission to infer it. - Aspect ratio, defaulting to vertical mobile composition. - One visual style from [style-recipes.md](style-recipes.md), defaulting to Tech White. - Whether a cover and final takeaway page are required. - Brand colors, logo, and mandatory copy, if any. ## Page planning 1. Extract claims, facts, steps, comparisons, and calls to action without changing their meaning. 2. Use roughly one page per 150 to 200 Chinese characters or three to four independent points. 3. Give every page one core message. Split crowded pages instead of shrinking text. 4. Build a narrative sequence: cover, context, core points, evidence or comparison, takeaway. 5. Keep exact statistics and quotations tied to the source. Do not invent supporting data. 6. Draft the complete page plan before any paid image call. Present a table with page number, title, purpose, exact visible copy, and visual device. Ask for approval of the page count, because every page is a separate generation call. ## Prompt assembly For each page, combine: - Shared series identity: ratio, style recipe, palette, typography direction, margins, and page marker. - Page content: title, subtitle, two to four short points, and exact data. - Layout: one suitable component such as bullet cards, number highlight, comparison, timeline, or quote. - Constraints: high whitespace, strong hierarchy, no invented words, and no decorative text. Repeat the invariant style description in every page prompt so the series does not drift. Keep visible copy short and warn that raster text may require regeneration or manual correction. ## Generation 1. Inspect `direct:fal_tools:fal_text_to_image` and calculate the current estimated total from the approved image count. 2. Obtain any required cost confirmation before the first call. 3. Generate the approved pages. Use conservative concurrency and stop the batch after a provider or billing failure. 4. Verify that every result contains an image asset before reporting success. 5. Return images in page order with titles and the exact prompt used for each. When restyling, keep the approved information architecture and change only the shared style recipe unless the user asks to rewrite content. ## Boundaries - Do not include private local paths, remote includes, or unlicensed brand assets. - Do not claim pixel-perfect text rendering. - Do not transform unsupported claims into facts or remove material source caveats. - Confirm rights for logos, photos, proprietary illustrations, and trademarked characters. ## Attribution This is a native Loadout re-authoring informed by the pinned source described in [UPSTREAM.md](UPSTREAM.md). The preserved source license is in [LICENSE.txt](LICENSE.txt). It removes the source's private paths and provider-specific local workflow.