← falCONTENT HISTORY

Update to fal

Snapshot Nov 14, 2026 · 00:10 UTC · version 1.3.2

Collection source: downloaded plugin package. These snapshots do not have a confirmed matching collection source. Differences in file lists alone do not establish changes to the package.

WHAT CHANGED

Instructions updated for fal-prompting

Skill instructions — Now says: Bundled guidance is fallback craft. For inputs, recommend_model's `input_contract` and live validation at submission are authoritative. See [the shared...

Observed in instructions or declared skills. Runtime behavior has not been tested.

Save fal and check future changes in Following.

Up to 3 plugins free · no card required.

All fields and original values.

Skill instructions

Before

Sep 30, 2026 · 22:46 UTC

model family, not generic "cinematic masterpiece" text. - Inspect schema before assuming negative prompts, seeds, multi-image inputs, multi-prompt arrays, duration, audio, or camera controls exist. Use for exact text, packaging, poster...

After

Nov 14, 2026 · 00:10 UTC

Bundled guidance is fallback craft. For inputs, recommend_model's `input_contract` and live validation at submission are authoritative. See [the shared production contract](../shared/production-contract.md). model family and actual task....

Compare saved observations

Download comparison JSON

SKILL.md text changes

− removed · + added

--- before
+++ after
@@ -8,23 +8,32 @@
 
 # fal.ai Prompting
 
+Bundled guidance is fallback craft. For inputs, recommend_model's
+`input_contract` and live validation at submission are authoritative. See
+[the shared production contract](../shared/production-contract.md).
+
 Use this after model selection. The goal is to write prompts that match the
-model family, not generic "cinematic masterpiece" text.
+model family and actual task. Family examples below are fallback craft; current
+published endpoint-specific prompting advice leads when you have it; fetch it
+through `search_guidance`/`read_guidance` only when the user asks or the prompt
+keeps failing, and retain its version. Use only control fields listed in
+`input_contract`.
 
 ## Universal Rules
 
 - Visual facts beat prestige adjectives.
 - Preserve one controlled variable per iteration.
-- Inspect schema before assuming negative prompts, seeds, multi-image inputs,
-  multi-prompt arrays, duration, audio, or camera controls exist.
+- Don't assume negative prompts, seeds, multi-image inputs, multi-prompt
+  arrays, duration, audio, or camera controls exist unless `input_contract`
+  lists them.
 - Keep exact text short and quoted when the model supports text.
 - For product and character work, name the invariant explicitly in every prompt.
 - For video, describe action over time, camera motion, framing, and ending.
 
 ## GPT Image 2
 
-Use for exact text, packaging, posters, UI, premium stills, and complex image
-composition.
+When this family has already been selected using live fit/guidance, a
+structured composition brief can include exact copy and packaging invariants.
 
 Prompt structure:
 
@@ -35,8 +44,8 @@
 5. Text requirements, only if needed.
 6. Guardrails: no extra logos, no unreadable text, preserve reference details.
 
-GPT Image 2 benefits from structured, explicit prompts. It can handle longer
-instructions than fast video models.
+Fallback pattern: structured, explicit instructions. Use the selected
+endpoint's current published prompting note for length and control guidance.
 
 ## Kling
 
@@ -55,7 +64,9 @@
 
 ## Happy Horse
 
-Use short natural-language prompts. Long production paragraphs usually hurt.
+Fallback example: a concise natural-language shot description. Use the
+current published prompting note for the selected endpoint before assuming
+that long prompts reduce its quality.
 
 Good pattern:
 
@@ -65,8 +76,7 @@
 
 ## Seedance
 
-Use for high-quality cinematic image-to-video and text-to-video. Build prompts
-like a short shot direction:
+After this family has been selected, a fallback shot direction includes:
 
 - Subject/action.
 - Camera motion.
@@ -81,8 +91,8 @@
 ## Prompt Failure Fixes
 
 - Product changed: switch to reference/edit workflow and tighten invariants.
-- Text is wrong: use GPT Image 2, shorten text, or reserve safe space for
-  external typography.
+- Text is wrong: inspect live task-specific guidance, shorten exact copy, or
+  reserve safe space for external typography. Do not switch families blindly.
 - Video drifts: reduce prompt scope, use image-to-video from an approved frame,
   shorten duration, or choose a model with stronger controls.
 - Character identity drifts: use the character anchor wording and reference
Full technical diff · 1 changed fields

changed /skill_md_contents

BEFORE
"---\nname: fal-prompting\ndescription: >\n  Apply model-family-specific prompting for fal.ai endpoints after an endpoint\n  has been selected. Use for GPT Image 2, Kling, Happy Horse, Seedance, product\n  prompts, exact text, multi-prompt video, and prompt debugging.\n---\n\n# fal.ai Prompting\n\nUse this after model selection. The goal is to write prompts that match the\nmodel family, not generic \"cinematic masterpiece\" text.\n\n## Universal Rules\n\n- Visual facts beat prestige adjectives.\n- Preserve one controlled variable per iteration.\n- Inspect schema before assuming negative prompts, seeds, multi-image inputs,\n  multi-prompt arrays, duration, audio, or camera controls exist.\n- Keep exact text short and quoted when the model supports text.\n- For product and character work, name the invariant explicitly in every prompt.\n- For video, describe action over time, camera motion, framing, and ending.\n\n## GPT Image 2\n\nUse for exact text, packaging, posters, UI, premium stills, and complex image\ncomposition.\n\nPrompt structure:\n\n1. Output type and composition.\n2. Exact subject/product/character.\n3. Style and medium.\n4. Lighting, material, camera, color.\n5. Text requirements, only if needed.\n6. Guardrails: no extra logos, no unreadable text, preserve reference details.\n\nGPT Image 2 benefits from structured, explicit prompts. It can handle longer\ninstructions than fast video models.\n\n## Kling\n\nUse for controlled video prompts when schema supports multi-prompt or element\ncontrols.\n\nPrompt structure:\n\n1. Subject identity and first frame.\n2. Shot type and camera motion.\n3. Action sequence in temporal order.\n4. Motion constraints.\n5. Ending state.\n\nKeep control fields schema-driven. Do not invent element fields without schema.\n\n## Happy Horse\n\nUse short natural-language prompts. Long production paragraphs usually hurt.\n\nGood pattern:\n\n`Handheld phone video, close shot of a runner tying neon shoes on wet pavement, soft morning light, quick push-in.`\n\nAvoid stacking style tags or long guardrails. Put important action first.\n\n## Seedance\n\nUse for high-quality cinematic image-to-video and text-to-video. Build prompts\nlike a short shot direction:\n\n- Subject/action.\n- Camera motion.\n- Environment.\n- Lighting.\n- Duration feel.\n- What should not change.\n\nFor image-to-video, the uploaded frame is the anchor. The prompt should animate\nthe frame, not redesign it.\n\n## Prompt Failure Fixes\n\n- Product changed: switch to reference/edit workflow and tighten invariants.\n- Text is wrong: use GPT Image 2, shorten text, or reserve safe space for\n  external typography.\n- Video drifts: reduce prompt scope, use image-to-video from an approved frame,\n  shorten duration, or choose a model with stronger controls.\n- Character identity drifts: use the character anchor wording and reference\n  images in every step.\n- Output is generic: replace style adjectives with concrete camera, light,\n  materials, setting, and action.\n"
AFTER
"---\nname: fal-prompting\ndescription: >\n  Apply model-family-specific prompting for fal.ai endpoints after an endpoint\n  has been selected. Use for GPT Image 2, Kling, Happy Horse, Seedance, product\n  prompts, exact text, multi-prompt video, and prompt debugging.\n---\n\n# fal.ai Prompting\n\nBundled guidance is fallback craft. For inputs, recommend_model's\n`input_contract` and live validation at submission are authoritative. See\n[the shared production contract](../shared/production-contract.md).\n\nUse this after model selection. The goal is to write prompts that match the\nmodel family and actual task. Family examples below are fallback craft; current\npublished endpoint-specific prompting advice leads when you have it; fetch it\nthrough `search_guidance`/`read_guidance` only when the user asks or the prompt\nkeeps failing, and retain its version. Use only control fields listed in\n`input_contract`.\n\n## Universal Rules\n\n- Visual facts beat prestige adjectives.\n- Preserve one controlled variable per iteration.\n- Don't assume negative prompts, seeds, multi-image inputs, multi-prompt\n  arrays, duration, audio, or camera controls exist unless `input_contract`\n  lists them.\n- Keep exact text short and quoted when the model supports text.\n- For product and character work, name the invariant explicitly in every prompt.\n- For video, describe action over time, camera motion, framing, and ending.\n\n## GPT Image 2\n\nWhen this family has already been selected using live fit/guidance, a\nstructured composition brief can include exact copy and packaging invariants.\n\nPrompt structure:\n\n1. Output type and composition.\n2. Exact subject/product/character.\n3. Style and medium.\n4. Lighting, material, camera, color.\n5. Text requirements, only if needed.\n6. Guardrails: no extra logos, no unreadable text, preserve reference details.\n\nFallback pattern: structured, explicit instructions. Use the selected\nendpoint's current published prompting note for length and control guidance.\n\n## Kling\n\nUse for controlled video prompts when schema supports multi-prompt or element\ncontrols.\n\nPrompt structure:\n\n1. Subject identity and first frame.\n2. Shot type and camera motion.\n3. Action sequence in temporal order.\n4. Motion constraints.\n5. Ending state.\n\nKeep control fields schema-driven. Do not invent element fields without schema.\n\n## Happy Horse\n\nFallback example: a concise natural-language shot description. Use the\ncurrent published prompting note for the selected endpoint before assuming\nthat long prompts reduce its quality.\n\nGood pattern:\n\n`Handheld phone video, close shot of a runner tying neon shoes on wet pavement, soft morning light, quick push-in.`\n\nAvoid stacking style tags or long guardrails. Put important action first.\n\n## Seedance\n\nAfter this family has been selected, a fallback shot direction includes:\n\n- Subject/action.\n- Camera motion.\n- Environment.\n- Lighting.\n- Duration feel.\n- What should not change.\n\nFor image-to-video, the uploaded frame is the anchor. The prompt should animate\nthe frame, not redesign it.\n\n## Prompt Failure Fixes\n\n- Product changed: switch to reference/edit workflow and tighten invariants.\n- Text is wrong: inspect live task-specific guidance, shorten exact copy, or\n  reserve safe space for external typography. Do not switch families blindly.\n- Video drifts: reduce prompt scope, use image-to-video from an approved frame,\n  shorten duration, or choose a model with stronger controls.\n- Character identity drifts: use the character anchor wording and reference\n  images in every step.\n- Output is generic: replace style adjectives with concrete camera, light,\n  materials, setting, and action.\n"
Full snapshot data
{
  "description": "Apply model-family-specific prompting for fal.ai endpoints after an endpoint has been selected. Use for GPT Image 2, Kling, Happy Horse, Seedance, product prompts, exact text, multi-prompt video, and prompt debugging.\n",
  "included_files": [],
  "name": "fal-prompting",
  "skill_md_contents": "---\nname: fal-prompting\ndescription: >\n  Apply model-family-specific prompting for fal.ai endpoints after an endpoint\n  has been selected. Use for GPT Image 2, Kling, Happy Horse, Seedance, product\n  prompts, exact text, multi-prompt video, and prompt debugging.\n---\n\n# fal.ai Prompting\n\nBundled guidance is fallback craft. For inputs, recommend_model's\n`input_contract` and live validation at submission are authoritative. See\n[the shared production contract](../shared/production-contract.md).\n\nUse this after model selection. The goal is to write prompts that match the\nmodel family and actual task. Family examples below are fallback craft; current\npublished endpoint-specific prompting advice leads when you have it; fetch it\nthrough `search_guidance`/`read_guidance` only when the user asks or the prompt\nkeeps failing, and retain its version. Use only control fields listed in\n`input_contract`.\n\n## Universal Rules\n\n- Visual facts beat prestige adjectives.\n- Preserve one controlled variable per iteration.\n- Don't assume negative prompts, seeds, multi-image inputs, multi-prompt\n  arrays, duration, audio, or camera controls exist unless `input_contract`\n  lists them.\n- Keep exact text short and quoted when the model supports text.\n- For product and character work, name the invariant explicitly in every prompt.\n- For video, describe action over time, camera motion, framing, and ending.\n\n## GPT Image 2\n\nWhen this family has already been selected using live fit/guidance, a\nstructured composition brief can include exact copy and packaging invariants.\n\nPrompt structure:\n\n1. Output type and composition.\n2. Exact subject/product/character.\n3. Style and medium.\n4. Lighting, material, camera, color.\n5. Text requirements, only if needed.\n6. Guardrails: no extra logos, no unreadable text, preserve reference details.\n\nFallback pattern: structured, explicit instructions. Use the selected\nendpoint's current published prompting note for length and control guidance.\n\n## Kling\n\nUse for controlled video prompts when schema supports multi-prompt or element\ncontrols.\n\nPrompt structure:\n\n1. Subject identity and first frame.\n2. Shot type and camera motion.\n3. Action sequence in temporal order.\n4. Motion constraints.\n5. Ending state.\n\nKeep control fields schema-driven. Do not invent element fields without schema.\n\n## Happy Horse\n\nFallback example: a concise natural-language shot description. Use the\ncurrent published prompting note for the selected endpoint before assuming\nthat long prompts reduce its quality.\n\nGood pattern:\n\n`Handheld phone video, close shot of a runner tying neon shoes on wet pavement, soft morning light, quick push-in.`\n\nAvoid stacking style tags or long guardrails. Put important action first.\n\n## Seedance\n\nAfter this family has been selected, a fallback shot direction includes:\n\n- Subject/action.\n- Camera motion.\n- Environment.\n- Lighting.\n- Duration feel.\n- What should not change.\n\nFor image-to-video, the uploaded frame is the anchor. The prompt should animate\nthe frame, not redesign it.\n\n## Prompt Failure Fixes\n\n- Product changed: switch to reference/edit workflow and tighten invariants.\n- Text is wrong: inspect live task-specific guidance, shorten exact copy, or\n  reserve safe space for external typography. Do not switch families blindly.\n- Video drifts: reduce prompt scope, use image-to-video from an approved frame,\n  shorten duration, or choose a model with stronger controls.\n- Character identity drifts: use the character anchor wording and reference\n  images in every step.\n- Output is generic: replace style adjectives with concrete camera, light,\n  materials, setting, and action.\n"
}

SHA-256 of public snapshot: baa1cf6774b9a410f5adb6082d7f6fcc52f43d060ace5554c8ec76c9e3d3add4