AUTO FLOW - SAMPLE PROMPT FORMATS

Use these as templates. Replace placeholder text with your own idea.

========================================
A) REGULAR PROMPTS (Text to Video / Create Image)
========================================

[V1-S1] Establishing wide shot of a rainy neon street at midnight, reflective puddles, cinematic depth, moody atmosphere.

[V1-S2] Medium shot of a courier character under a flickering sign, soaked jacket, breathing visible in cold air, tense expression.

[V1-S3] Close-up of gloved hand gripping a glowing data chip, droplets sliding across metal surface.

[V1-S4] Over-the-shoulder shot entering a narrow alley, holographic ads flicker, camera slowly pushes forward.

[V1-S5] Final hero shot on rooftop at dawn, storm fading, city lights transitioning to sunrise.

Tips:
- One prompt per block (blank line between prompts) keeps sorting cleaner.
- Keep tags like [V1-S1] to preserve sequence.

========================================
B) IMAGE -> VIDEO WORKFLOW (Frame to Video)
========================================

Format:
[V#-S#] image prompt ||| [V#-S#] video prompt

Example:
[V1-S1] Cyberpunk city skyline at night, heavy rain, blue-magenta neon reflections ||| [V1-S1] Camera tilts down from skyline to street level as rain intensifies and lights flicker.

[V1-S2] Courier standing near glowing vending machine, steam vents in background ||| [V1-S2] Slow dolly-in toward courier while coat fabric moves in wind and steam drifts across frame.

[V1-S3] Macro shot of wet data chip in hand, etched circuitry glowing amber ||| [V1-S3] Extreme close-up rack focus from raindrops to chip core as light pulses.

[V1-S4] Alley entrance with hologram signs and wet asphalt ||| [V1-S4] Handheld-style forward movement through alley with slight shake and parallax.

[V1-S5] Rooftop sunrise reveal over megacity, storm clouds breaking ||| [V1-S5] Wide crane-up shot as clouds part and warm sunlight spreads over skyline.

[V2-S1] Desert outpost at golden hour, dust in the air, long shadows ||| [V2-S1] Slow panoramic sweep left to right revealing outpost scale.

[V2-S2] Pilot checking damaged drone wing near hangar ||| [V2-S2] Medium orbit shot around pilot while dust particles move through sunbeams.

[V2-S3] Close-up of cracked visor reflecting incoming sandstorm ||| [V2-S3] Push-in close-up with subtle head turn and reflected lightning in visor.

[V2-S4] Wide shot of convoy leaving outpost, engines kicking dust ||| [V2-S4] Tracking side shot matching convoy speed with dynamic motion blur.

[V2-S5] Night camp under stars with blue emergency lights ||| [V2-S5] Slow pullback from campfire to wide aerial as lights fade into darkness.

Tips:
- Keep the same scene tag on both sides of ||| for clarity.
- Left side = image generation prompt.
- Right side = motion/animation prompt for that same scene.
- Use one row per final video.

Multiple images into one video:
[V1-S6] @img1 [0.0-3.0s] wide shot of @host at the gym bench ;; @img2 [3.0-6.0s] close-up of @host picking up the product ;; @img3 [6.0-9.0s] insert shot of the product in hand ||| [V1-S6] Use @img1, @img2, and @img3 as timed inputs for one final video. Do not make one video per image.

Multi-frame rules:
- Keep all still image requests on the left side of |||.
- Separate @img frames with ;;.
- The right side is always the single final video prompt for that row.

Agent Mode AFID tracking:
- Agent Mode builds an expected manifest from each row tag.
- Example row:
  [V1-S10] image1 ;; image2 ;; image3 ||| [V1-S10] video
- Expected image slots:
  AFID:V1-S10-IMG1
  AFID:V1-S10-IMG2
  AFID:V1-S10-IMG3
- Final video ID: [V1-S10]
- Expected image count: 3
- Expected video count: 1
- Preserve AFID in generation prompt metadata for tracking.
- Do not render AFID, labels, captions, or text in the image/video.

Copyable Agent Mode packet:
# AGENT INSTRUCTIONS
Use one row per final video. Preserve [V#-S#] row tags and AFID strings in prompt metadata. Do not render AFID, labels, captions, or visible text.

# PROJECT BRIEF
Describe the product, characters, style, reference rules, dialogue rules, and QA bar for this run.

# REFERENCES
List the uploaded references and which rows, @handles, or image slots they should guide.

# VIDEO PROMPTS
[V1-S1] AFID:V1-S1-IMG1 still image prompt ;; AFID:V1-S1-IMG2 optional second still ||| [V1-S1] final frame-to-video prompt using those stills.

# QA EXPECTATIONS
Check dialogue, speaker, timing, identity, product accuracy, and whether the ending is complete.

# RETRY ROWS
Retry only the listed row IDs, keep unrelated rows unchanged, and state the issue to fix. CLI/MCP controls are coming soon.

Agent Mode in plain English:
- Flow Agent can create inside Flow. Auto Flow Agent Mode adds the operating layer around it: row IDs, AFID tracking, retry scope, reconcile reports, download plans, QA notes, and external agent control.
- Use one row per final video. If a row needs multiple stills first, keep them on the left side of ||| separated by ;;.
- Keep [V#-S#] row tags stable. They are how Auto Flow knows what should exist, what is missing, and what should be retried.
- AFID strings are tracking metadata. They must not appear as visible text in the generated image or video.

One-image / one-video smoke packet:
# AGENT INSTRUCTIONS
Run the smallest possible Agent Mode smoke. Produce one still image and one final video. Preserve row tags and AFID metadata. Do not add visible labels, captions, or AFID text.

# PROJECT BRIEF
Simple product-style test shot. Clean studio lighting. No dialogue.

# VIDEO PROMPTS
[V1-S1] AFID:V1-S1-IMG1 a single matte black ceramic mug on a walnut desk, soft morning window light, realistic product photo ||| [V1-S1] slow 4 second push-in on the mug, subtle light movement, stable product identity, no text overlays

# QA EXPECTATIONS
Confirm exactly one expected image slot and one final video. Check that the mug stays consistent, no AFID or labels are visible, and the ending is complete.

QA transcript workflow:
- Export or paste the final transcript after generation.
- Review row by row against # QA EXPECTATIONS.
- Mark each row PASS, RETRY, or BLOCKED.
- For RETRY, name the row ID and the specific defect, for example wrong speaker, clipped ending, missing product, bad timing, identity drift, unsafe visible text, or bad audio.
- Retry only the listed rows. Do not rewrite passed rows.

Lane cleanup blockers:
- Stop and clean the lane if the mounted extension id is unknown, the Flow tab is not connected, Auto Flow says sign in is required, the project id cannot be read, duplicate Flow tabs make ownership ambiguous, or the sidepanel is pointed at a stale dist.
- Do not call a generation failure when the blocker is lane/auth/setup. Fix the lane first, then rerun the smoke.

CLI/MCP handoff for external coding agents:
- Use CLI/MCP for status, reconcile, download planning, and explicit guarded submit/retry commands.
- Read-only calls should work without sidepanel focus when the bridge is connected.
- Submit/retry must require an explicit row/run payload and should return a traceable request id.
- If the bridge is unavailable, report BLOCKED with the port, extension id, and missing connection detail instead of guessing.

========================================
C) CHARACTER PROMPTS
========================================

Basic:
@name character description

Advanced:
@name character description ||| voice: Flow voice name ||| info: extra character notes

Examples:
@maya Maya is a cyberpunk courier with silver hair. ||| voice: Achernar ||| info: Speaks in short sentences and avoids slang.

@rao Rao is a tired detective in a white trench coat. ||| voice: Charon

@drone_dog DroneDog is a small robotic companion. ||| voice: none ||| info: Communicates with chirps and movement.

Rules:
- Start each character line with @name.
- No spaces inside @name.
- Voice is optional.
- Voice must match Flow's fixed voice list.
- info is optional.
- Use the same @name in scene/video prompts.
- Undefined @mentions are highlighted when Character Prompts are active.
- Character creation does not count against prompt quota.

Existing Flow Characters:
- If the character already exists in the Flow project, keep the same @handle in
  scene prompts and map it in Character Mapping.
- Display names with spaces can map to underscore handles, for example
  "gary before" can map to @gary_before.
- For long Omni scripts, bind only the important speaker/identity mentions
  first. Binding every repeated @name is not yet a safe contract.

========================================
D) AUTO FLOW PACKET IMPORT
========================================

Use Paste Auto Flow Packet when you want to paste characters and video prompts
in one block. Auto Flow previews the packet, then applies the Character Prompts
and Video Prompts into the existing sections.

Format:
# CHARACTER PROMPTS
@test_luma Luma Test Character is a fictional young adult explorer with a teal jacket, silver backpack, short black hair, and a warm expression. Not based on any real person. ||| voice: none ||| info: curious, calm, adventurous

@test_orren Orren Test Character is a fictional young adult inventor with an amber hoodie, round glasses, curly brown hair, and a thoughtful expression. Not based on any real person. ||| voice: none ||| info: clever, nervous, precise

# VIDEO PROMPTS
[V1-S1] @test_luma stands at the edge of a glowing forest path, teal jacket catching soft blue light, cinematic fantasy realism. ||| Slow dolly-in as @test_luma looks toward the trees, curious and cautious.

[V1-S2] @test_orren kneels beside a small mechanical device on a wooden table, amber hoodie sleeves rolled up, warm workshop lighting. ||| Camera pushes in as @test_orren adjusts the device and glances toward @test_luma.

Copy this prompt into your LLM to generate an Auto Flow Packet:
You are writing an Auto Flow Packet. Use only # CHARACTER PROMPTS and # VIDEO PROMPTS. Define every @handle before using it. Use fictional non-famous characters. Use lowercase snake_case handles. Do not invent undefined @mentions. Use exactly one ||| split per video row. Left side = image or still-frame request. Right side = one final video request. For multiple images into one video, put @img1, @img2, and @img3 on the left side separated by ;;, then tell the video side to use those timed inputs for one final video. Preserve AFID in generation prompt metadata for tracking. Do not render AFID, labels, captions, or text in the image/video.

Rules:
- Packet import does not make Omni ready by itself.
- Omni/Path C stays character_setup_required until Flow native Character rows,
  AT_TAG_TYPE, and characterServerId proof exist.
- After applying a packet, use Create / Verify Native Characters. Auto Flow
  verifies existing mappings or creates missing native Flow Characters from
  the Character Prompts, then stores @handle -> characterServerId mappings
  only after true Character row proof exists.
- Source images/project grid image assets count only as
  character_asset_created_but_not_native_character. They are ready for Auto
  Flow Classic Match mapping, not Omni/Path C submission.
- Native setup may pass through character_character_row_pending while Flow
  finishes surfacing the Character row in the @ picker.
- Saved character mappings are preserved.

========================================
E) MATCHING MODES
========================================

Flow Native @ Match:
- Google's native character binding for Flow/Omni.
- Produces real Flow character chips and entity references.
- Use this for existing Flow characters and character videos.

Auto Flow Classic Match:
- Auto Flow's original reference matcher.
- Matches uploaded image refs to prompt rows by @name or filename.
- Works for ref-using modes such as Create Image, Frame to Video, and
  Ingredients to Video. It is not Omni-only.
