Do not share the studio access token, provider keys, private reference images, or generated links outside the approved team.
VA OPERATING GUIDE
Make strong content.
Know every check.
This guide explains what each AIF Studio area does, when it costs generation credits, and what must be checked before any file is posted.
Fast routeChoose a tool→Generate or process→Review every output→Run Metadata Checker→Download and post
READ FIRST
Rules that apply everywhere
Her viewer left iris is muted natural sage green. Her viewer right iris is warm medium brown. Keep her freckles, warm golden brown complexion, approved long dark wavy hair, and recognisable face.
Never copy the source person's face, skin, hair style, tattoos, piercings, jewellery, age cues, or identifying traits. Copy Outfit preserves garments only.
HART and U.S. ARMY appear only on a complete long sleeve OCP blouse and matching duty uniform. Tank tops, gym wear, sleepwear, casual clothes, and mixed OCP outfits have no clothing text.
TAB 1
Image Studio
Use this for new single images from a written scene.
- Amaya is selected automatically.
- Choose the correct content mode.
- Use Surprise Me, a suggestion, or write the scene yourself. Like good bank prompts and skip weak ones.
- Choose shape and image count. Fine Tune controls exclusions, seed, realism strength, and the SFW model.
- Generate, then inspect every result at full size before downloading.
Best use: new selfies, lifestyle pictures, fashion images, and Army scenes.
Costs credits: generation and provider based refinements.
Local and free: downloads and browser previews.
TAB 2
Carousel Generator
Use this for a related set of three to six images.
- For 5 Poses, upload one approved seed image. The seed controls outfit, setting, and composition.
- Army Selfies and Scenes needs no seed. It creates a selfie led mix across varied Army lifestyle settings.
- Story Generator needs no seed. It mixes three Amaya selfies with two person free lifestyle pictures in each five image story.
- Choose image count, model, and an image size supported by that model.
- Approve the useful frames. Use the Lighting menu for a local realism adjustment. Use the standalone Caption Editor for the largest caption library.
Recovery: if one image fails, the set can replace that slot. Do not start a second paid set while replacements are still running.
Lighting: local adjustments do not spend generation credits.
TAB 3
Viral Caption Editor
Use this to burn TikTok style text onto two to six existing images.
- Add images in posting order. You can return and add more later. New files append instead of replacing the existing set.
- Use the small × on any thumbnail to remove it. Use Randomise picture order when you want a fresh sequence.
- Choose the closest caption library, the number of variations, and the text position. Middle is the normal starting position.
- Build the sequence, edit any line that feels unnatural, and preview the chosen variant.
- Download Variant ZIPs. Each caption version has its own ZIP inside the master download.
Voice: captions use natural lowercase Gen Z phrasing and light shorthand such as u or smthng. Remove any wording that feels forced for the picture.
Local and free: caption rendering, reordering, previews, and ZIP creation happen in the browser.
TAB 4
Metadata Checker
Use this as the last check for every image batch.
- Add any number of final downloaded images. You can add more without losing the current list.
- Read the result for each file. A pass checks JPEG format, Apple camera fields, a believable capture date, and Houston, Texas location.
- If fields are missing, choose Add metadata to missing.
- Recheck the repaired downloads and use only files that pass.
Privacy: checking and repair happen locally in the browser. The uploaded files are not sent to a generation provider.
Date rule: reject obviously incorrect creation dates such as 1979.
SEPARATE PAGE
Recreate Video
Use this to rebuild a reference image or the first fifteen seconds of a video as Amaya.
- Upload a photograph or video. Longer videos are trimmed to the first fifteen seconds.
- Choose SFW Gemini for ordinary material. Choose NSFW Grok only when the source requires adult visual analysis.
- Choose the starting frame model and dimensions. Turn Copy Outfit on when the source garments must stay the same.
- Decode the source. Review the reconstruction summary and edit the generated prompt before spending credits on variants.
- Generate four variants, choose one, then use only the needed refinements. Colour swap changes top, bottom, or both. No Clothing Text removes garment lettering.
- Use motion transfer only after selecting a completed AIF generation. The Seedance production pack prepares the prompt and references for manual Unlimited generation.
Important: Copy Outfit never imports reference hair, tattoos, piercings, jewellery, or identity.
Speed: representative frame sampling is the normal route. Analyse Every Frame is much slower and should be used only when tiny actions are genuinely being missed.
SEPARATE PAGE
Viral Outliers
Use this to turn Handler research into an actionable recreation brief.
- Run or open the Handler research and paste the research URL into AIF.
- Review the hook, payoff, comments, caption pattern, and reusable creative structure.
- Send the chosen brief to Recreate. Upload the original source media there when the workflow needs visual evidence.
- Copy the structure, not the creator's identity or personal details.
Subscription: if Handler limits research volume, use fewer high quality examples first. Upgrade only when the current limit is the proven bottleneck.
MANDATORY
Final review before posting
TROUBLESHOOTING
What common messages mean
Insufficient credits
The external provider rejected the request because its wallet is empty. Do not keep retrying. Tell the account owner which provider is named in the error.
Failed to fetch
Keep the page open, check the connection, and retry once. If it repeats, copy the exact message and report the tab, model, action, and approximate time.
Safety rejection
The selected provider would not process the request. Confirm the correct content mode and use the approved workflow. Do not disguise the request or repeatedly submit it.
Output looks wrong
Do not post it. Save a screenshot, note the model and prompt, then report the exact defect, for example wrong eyes, copied tattoos, wrong outfit, weak action, duplicate person, or incorrect dimensions.
Download is unavailable
Wait until the result says Completed and the full image is visible. A queued, running, failed, or preview only item is not downloadable.