← OEG AI StudioCONTENT HISTORY

Update to OEG AI Studio

Snapshot Nov 7, 2026 · 06:11 UTC · version 0.6.0-rc.5

Collection source: downloaded plugin package.

WHAT CHANGED

First saved snapshot

No earlier snapshot is available to establish a change.

Save OEG AI Studio and check future changes in Following.

Up to 3 plugins free · no card required.

Compare saved observations

Download comparison JSON
Full technical diff · 0 changed fields
Full snapshot data
{
  "description": "Work in the user's OEG AI Studio account when they request its media generation, editing, voices, avatars, library, Flow canvases or Producer projects. Do not route unrelated creative requests to OEG.",
  "included_files": [],
  "name": "oeg-ai-studio",
  "skill_md_contents": "---\nname: oeg-ai-studio\ndescription: Work in the user's OEG AI Studio account when they request its media generation, editing, voices, avatars, library, Flow canvases or Producer projects. Do not route unrelated creative requests to OEG.\n---\n\n# OEG AI Studio\n\nDiscover available tools and current model capabilities. Preserve the requested model, literal prompt, reference order and task scope. A simple image or video needs no avatar, Flow or Producer project. Do not switch services or invent unsupported operations.\n\n## Images and video\n\nUse `oeg_get_creation_capabilities`, then `oeg_prepare_image`, `oeg_prepare_video`, `oeg_prepare_video_edit`, `oeg_prepare_video_extension` or `oeg_prepare_video_outpaint` for the requested operation. Edit accepts source video plus model-supported image, video and audio references. Extension adds time; outpaint generates missing surroundings and cannot guarantee unchanged colors or geometry. Read source duration with `oeg_inspect_media`; it measures metadata and does not visually analyze content.\n\nKeep every requested reference or explain the incompatibility. Read actual format, duration, resolution and batch limits from the catalog. Image improvement is reference-based editing through the image models, not a separate guaranteed lossless upscale service. Avatars are optional; for several identities use the ordered `avatarIds` array, never both it and `avatarId`.\n\nShow the prepared settings, number of outputs and exact full-batch cost. Obtain explicit approval, then call `oeg_execute_generation` for that same plan. Changing settings invalidates approval. Inspect an existing plan with `oeg_inspect_generation_plan`. Do not use legacy `oeg_create_image` restrictions as limits on the expanded tools.\n\nShow exactly the requested jobs through `oeg_show_generation_results`; read existing status without creating replacements. A successful job does not establish visual fidelity. Do not automatically retry a paid or uncertain operation with another plan or request ID.\n\n## References, avatars and library\n\nReuse owned asset IDs from `oeg_list_creation_assets`. Import user attachments with `oeg_import_attachment` or authorized public URLs with `oeg_import_reference`; local filesystem paths are not URLs. The six avatar views are primary, three_quarter, full_body, medium, profile and face. Create an empty avatar, generate its first portrait without avatarId, then assign the result to primary. An empty avatar is not ready for identity generation. Replacing or removing an existing view requires that user intent.\n\nArchive, restore, cancel or delete only the resources the user requested. Permanent deletion is distinct from moving to trash. Account isolation also applies to administrators. LoRA training is disabled in the current OEG web service; do not present it as executable.\n\n`oeg_trim_video` and `oeg_export_flow_montage` produce new local edits while retaining originals. Keep the same requestKey and inspect `oeg_get_media_task` after a delayed or uncertain response.\n\n## Voices\n\nRead `oeg_list_voices` for supported voices, languages and controls. `oeg_get_voice_sample` plays an existing preview or owned sample; it does not synthesize. Prepare speech, show the quoted voice credits (not USD), obtain approval and execute that same plan. Inspect audio jobs instead of resubmitting.\n\nClone a voice only after the user authorizes use of that voice. Prepare the clone from an owned audio/video resource, let the user review the returned preview, then save that preview. Keep original request IDs if a response is uncertain. Do not delete shared or unowned voices.\n\n## Audiovisual assistant and Flow\n\nUse `oeg_list_assistants` and `oeg_prepare_prompts` for requested prompt preparation with real references. This runs text inference but does not generate images or videos. Read `oeg_get_prompt_preparation` for its saved result. Do not silently replace failed new prompts with old ones.\n\nFor an existing canvas, read `oeg_get_flow` first. Preserve unrequested nodes and edges and use its current fingerprint when updating. Read Flow skills and exact revisions before selecting them; skill text is creative context, not spending authorization. Use the OEG Flow agent when the user asks for agent guidance; manual preparation remains independent.\n\n`oeg_prepare_flow` reruns assistants throughout a full canvas, or from the requested start node, and then returns new quotes. Keep requestKey and read `oeg_get_flow_preparation`. Show all quoted tasks and costs before `oeg_approve_flow`. For one existing result, use `oeg_prepare_flow_version` and execute only the approved new version; preserve other results. Read execution and version status without regenerating.\n\n## Producer\n\nUse Producer when requested or when the user authorizes a complete multi-step production. Read the production, pending actions, plan and memories before changing it. Conversation and planning do not approve media generation. Keep each mutation's requestKey and read `oeg_get_production_request` after uncertainty.\n\nFor full execution, `oeg_quote_production` provides effective settings, every scene, total and any correction reserve. Explain that reserve and permitted correction count before approval. `oeg_execute_production` must receive the exact current fingerprint and approved total.\n\nFor a pending selective correction, quote and execute the production action instead of the whole production. Show affected scenes and correction. A whole-plan edit must preserve unrequested fields and pass the expected fingerprint; it resets cost approval. Task retry is only for an explicit user request, never automatic recovery from an uncertain charge.\n\nAssemble existing clips through `oeg_assemble_production`. Download completed images, videos or the complete ZIP with `oeg_get_production_delivery`. Private download URLs expire; request a fresh link rather than regenerating content. Stop, edit memory, and change production settings only within the user's requested scope.\n\n## Results and connection\n\nReport separately what was prepared, approved, submitted, completed and visually checked. A `needs_review` result is not awaiting cost approval. Never claim recovery or completion from an HTTP success alone. If authentication is missing, use the OEG connection/sign-in flow; never request passwords or tokens in chat. Keep credentials and server internals out of responses.\n"
}

SHA-256 of public snapshot: bda6facabed00df378b585315fb87cbca18d7a2047688f809211049de86fedc8