Skip to content

Provider prompting guide tracker

This tracker is for contributors who maintain model-specific examples. Use it to find provider sources, update the offline handbook, and record recipe validation. For instructions on reading guides, see Cookbooks.

Inventory snapshot: September 4, 2026, repository commit b4a692ec144f2e5f3c61504e657dbbb1989c6c09. The table accounts for all 139 managed model IDs in the canonical catalog, grouped into 59 rows. Each exact ID appears once. This inventory excludes aliases, separately managed adapters, and hidden companion artifacts.

Status and coverage

Source discovery and local recipe validation have separate statuses. Table 1 defines the source statuses and records their coverage.

Table 1. Source status definitions and coverage
Source statusMeaningManaged IDs
GuideA provider-authored prompting guide, cookbook, or prompt-preparation resource was retrieved. Applicability can still be family-level.38
ReferenceA provider model card, repository, format reference, or examples entry point was retrieved. A dedicated guide and detailed recipe extraction remain open.57
Research pendingProvider guidance has not been verified in this pass. This doesn't mean no guide exists.12
No free-text taskThe managed entry is a component or task-specific input pipeline. Track input preparation and conditioning instead of a generative prompt recipe.32

All 59 local recipes have Drafted status and are bundled for offline reading. To read a recipe in macOS Studio, in Help, select mere.run Guide. Each recipe includes original examples and distinguishes retrieved provider material from local workflow advice. The handbook has no recorded inference validation. Source research gaps remain separate from recipe validation.

Provider documentation can describe controls that the local runtime doesn't expose. Before applying provider advice, check the exact checkpoint, chat format, input type, and local command. Quantization reduces weight precision and can change results. Fine-tuning, distillation, and hosted prompt rewriting can also affect how a recipe transfers.

The Checked column records source retrieval for Guide and Reference rows. "Not verified" means that no provider source was verified. Source links point to provider documentation, repositories, or model cards. Local recipe links point to the bundled Markdown files in this repository.

Image

Table 2 lists provider sources and local recipes for image models.

Table 2. Image model guide coverage
Family and providerManaged model IDsSource status and linksScope and next actionLocal recipeChecked
FLUX.2 Klein (Black Forest Labs)image-klein-nano, image-klein-max, image-klein-9b, image-klein-base, image-klein-base-9bGuide. FLUX prompting guideFamily guidance; map base versus distilled settings and local reference-image controls.FLUX.2 Klein (Black Forest Labs) guide (draft)2026-09-04
FLUX.2 dev (Black Forest Labs)image-flux2-devGuide. FLUX prompting guideUse the dev-relevant sections; hosted pro and max controls require separate compatibility review.FLUX.2 dev (Black Forest Labs) guide (draft)2026-09-04
FLUX.1 dev (Black Forest Labs)image-flux1-devGuide. FLUX prompting guideThe guide explicitly covers FLUX.1; extract generation examples for dev.FLUX.1 dev (Black Forest Labs) guide (draft)2026-09-04
Klein shared componentsimage-klein-sharedNo free-text taskComponent bundle; track prompting with the consuming Klein model.Klein shared components guide (draft)Not verified
Bonsai image (PrismML)image-bonsai-binary, image-bonsai-ternaryResearch pendingFind the publisher's binary and ternary guidance; don't assume another image family's prompt format.Bonsai image (PrismML) guide (draft)Not verified
Z-Image (Tongyi-MAI)image-zimage-nano, image-zimage-max, image-zimage-baseReference. Z-Image repositoryGeneration examples; extract separate base and Turbo recommendations.Z-Image (Tongyi-MAI) guide (draft)2026-09-04
HiDream O1 (HiDream-ai)image-hidream-o1, image-hidream-o1-devReference. HiDream O1 model cardReview prompt preparation and the Dev card separately before sharing one recipe.HiDream O1 (HiDream-ai) guide (draft)2026-09-04
SenseNova U1.5 (SenseNova)image-sensenova-u1-5-8b-motGuide. SenseNova U1.5 cookbookExact family cookbook; map generation, editing, and optional prompt enhancement.SenseNova U1.5 (SenseNova) guide (draft)2026-09-04
Krea 2 (Krea)image-krea2-raw, image-krea2-turboGuide. Krea 2 prompting guideHosted-product guide; verify which exploration and style-reference techniques transfer to Raw and Turbo.Krea 2 (Krea) guide (draft)2026-09-04
Qwen Image Edit 2511 (Qwen)image-qwen-edit-2511, image-qwen-edit-2511-lightningReference. Qwen Image Edit 2511 model cardExact base release examples; Lightning needs its own adapter and settings review.Qwen Image Edit 2511 (Qwen) guide (draft)2026-09-04
Ideogram 4 (Ideogram)image-ideogram4-sdnq-uint4Research pendingLocate provider guidance for version 4 and distinguish it from the WaveCut quantization card; attempted guide URL could not be retrieved.Ideogram 4 (Ideogram) guide (draft)Not verified

Text and multimodal

Table 3 lists provider sources and local recipes for text and multimodal models.

Table 3. Text and multimodal model guide coverage
Family and providerManaged model IDsSource status and linksScope and next actionLocal recipeChecked
MeBot and Psitext-chat-mebot, text-chat-psi-agentResearch pendingCatalog entries have no managed Hub source; identify checkpoint provenance before assigning provider guidance.MeBot and Psi guide (draft)Not verified
Gemma 4 (Google)text-chat-gemma4, text-chat-gemma4-turbo, text-chat-gemma4-12b, text-chat-gemma4-12b-4bit, vision-chat-gemma4-12b, text-chat-gemma4-nano, text-chat-gemma4-maxReference. Gemma 4 prompt formattingOfficial roles, control tokens, and multimodal format; verify each checkpoint's generation settings.Gemma 4 (Google) guide (draft)2026-09-04
DiffusionGemma (Google)text-chat-diffusiongemma-26b-optiq-4bitResearch pendingFind diffusion-specific prompting instructions; Gemma 4 formatting alone doesn't establish compatibility.DiffusionGemma (Google) guide (draft)Not verified
Laguna 2.1 (poolside)text-chat-laguna-s-2-1, text-chat-laguna-xs-2-1Reference. Laguna S 2.1 model cardSource located. Inspect linked base-model instructions and verify XS separately.Laguna 2.1 (poolside) guide (draft)2026-09-04
Inkling Small (Thinking Machines)text-chat-inkling-smallReference. Inkling Small model cardSource located. Extract exact chat, reasoning, and tool conventions.Inkling Small (Thinking Machines) guide (draft)2026-09-04
Muse Glimmer (Meta)vision-chat-muse-glimmer-30bReference. Muse Glimmer model cardSource located. Review multimodal and conversation formatting for the managed quantization.Muse Glimmer (Meta) guide (draft)2026-09-04
Nemotron 3.5 Lightning (NVIDIA)text-chat-nemotron-35-lightningReference. Lightning model cardExact upstream release; extract reasoning and sampling guidance.Nemotron 3.5 Lightning (NVIDIA) guide (draft)2026-09-04
Nemotron 3 Nano Omni (NVIDIA)omni-chat-nemotron3-nano-30b-a3b-bf16Reference. Omni model cardExact upstream release; map modality-specific examples to local supported inputs.Nemotron 3 Nano Omni (NVIDIA) guide (draft)2026-09-04
Qwen 3.5, 3.6, and 3.8 (Qwen)text-chat-q36-nano, vision-chat-q38-27b, vision-chat-q38-27b-4bit, vision-chat-q38-flash-next-mixed, vision-chat-q38-flash-next-3bit, vision-chat-q38-flash-next-3bit-native-ple, vision-chat-q38-flash-next-4bit, text-agent-qwen35-9b, text-chat-q36-nano-ggufReference. Official series repository; Qwen cookbookSeries entry point; follow each exact model card. Track 3.8 Flash-Next separately from 27B when writing recipes.Qwen 3.5, 3.6, and 3.8 (Qwen) guide (draft)2026-09-04
Bonsai text (PrismML)text-chat-bonsai-27b-1bit, text-chat-bonsai-27b-2bitResearch pendingFind publisher prompting and generation settings for each low-bit checkpoint.Bonsai text (PrismML) guide (draft)Not verified
Ornith 1.5 (Ornith)text-agent-ornith-35b-mlx-4bit, text-agent-ornith-35b-mlx-6bit, text-agent-ornith-35b-mlx-8bit, text-agent-ornith-35b-mlx, vision-chat-ornith-35bReference. Ornith 1.5 model cardFamily source; map text, vision, tool templates, and quantized variants.Ornith 1.5 (Ornith) guide (draft)2026-09-04
Ornith 1.0text-agent-ornith-9b, text-agent-ornith-35bResearch pendingLocate 1.0 publisher instructions for 9B and 35B; don't inherit 1.5 guidance silently.Ornith 1.0 guide (draft)Not verified
North Mini Codetext-code-north-miniResearch pendingResolve original publisher from the managed GGUF artifact and locate its prompting instructions.North Mini Code guide (draft)Not verified
DeepSeek V4 Flash (DeepSeek)text-agent-deepseek-v4-flashResearch pendingFind V4-specific provider guidance and compare it with the managed GGUF chat template.DeepSeek V4 Flash (DeepSeek) guide (draft)Not verified
LFM2.5 text (Liquid AI)text-chat-lfm25-a1b-8bit, text-chat-lfm25-a1b-bf16, text-chat-lfm25-1.2b-bf16, text-chat-lfm25-1.2b-qad-4bit, text-chat-lfm25-2.6b-4bit, text-chat-lfm25-2.6b-qad-4bit, text-chat-lfm25-2.6b-bf16Guide. Prompting guideReview roles, input and output examples, and response prefixes. Use exact model cards for sampling settings.LFM2.5 text (Liquid AI) guide (draft)2026-09-04
LFM2.5 vision (Liquid AI)vision-chat-lfm25-3b-8bitGuide. Vision capabilitiesIncludes VL-3B examples for image questions, grounding, layout, and tools; map to local controls.LFM2.5 vision (Liquid AI) guide (draft)2026-09-04
Qwen3 code (Qwen)text-code-qwen3Reference. Qwen3-Coder repositoryFamily examples; confirm exact checkpoint before adopting tool formatting or sampling defaults.Qwen3 code (Qwen) guide (draft)2026-09-04
Qwen3 text embeddings (Qwen)text-embed-qwen3-0.6bReference. Embedding usageExtract retrieval task instructions and query-versus-document formatting.Qwen3 text embeddings (Qwen) guide (draft)2026-09-04
Qwen3 VL embeddings (Qwen)vision-embed-qwen3-vl-2bReference. VL embedding usageExtract query instructions and multimodal input examples.Qwen3 VL embeddings (Qwen) guide (draft)2026-09-04
Privacy Filtertext-anonymize-privacy-filterNo free-text taskTask-specific anonymization; track input and entity controls rather than a generative prompting recipe.Privacy Filter guide (draft)Not verified

Speech

Table 4 lists provider sources and local recipes for speech models.

Table 4. Speech model guide coverage
Family and providerManaged model IDsSource status and linksScope and next actionLocal recipeChecked
Qwen3 TTS (Qwen)speech-tts-qwen3-nano, speech-tts-qwen3-customvoiceReference. Qwen3-TTS usageSeparate VoiceDesign descriptions from CustomVoice delivery instructions and spoken text.Qwen3 TTS (Qwen) guide (draft)2026-09-04
Qwen3 ASR (Qwen)speech-asr-qwen3Reference. Qwen3-ASR usageReview context and language conditioning; distinguish transcription context from generative instructions.Qwen3 ASR (Qwen) guide (draft)2026-09-04
Parakeet and Sortformer (NVIDIA)speech-asr-parakeet, speech-diarization-sortformerNo free-text taskTranscription and speaker diarization; track audio requirements and task controls.Parakeet and Sortformer (NVIDIA) guide (draft)Not verified

Vision and reconstruction

Table 5 lists provider sources and local recipes for vision and reconstruction models.

Table 5. Vision and reconstruction model guide coverage
Family and providerManaged model IDsSource status and linksScope and next actionLocal recipeChecked
Infinity Parser2 Pro (infly)vision-ocr-infinity-pro, vision-ocr-infinity-pro-int8Reference. Parser2 Pro model cardExtract document parsing prompt and output format; verify the Int8 variant separately.Infinity Parser2 Pro (infly) guide (draft)2026-09-04
LightOnOCR 2 (LightOn)vision-ocr-lightonReference. LightOnOCR model cardReview fixed document-recognition input and template conventions before suggesting user-written prompts.LightOnOCR 2 (LightOn) guide (draft)2026-09-04
SAM 3.1 (Meta)vision-segment-sam31Reference. SAM repository and notebooksFamily examples for text and geometric prompts; verify SAM 3.1-specific differences.SAM 3.1 (Meta) guide (draft)2026-09-04
Falcon Perception (TII)vision-ground-falcon-perceptionReference. Falcon Perception model cardExtract referring-expression and multi-object query conventions.Falcon Perception (TII) guide (draft)2026-09-04
TerraMind fire and floodvision-flood-terramind-base, vision-fire-terramind-baseNo free-text taskManaged task heads consume geospatial imagery; track bands and preprocessing in the runtime guide.TerraMind fire and flood guide (draft)Not verified
TESSERA and OlmoEarth embeddingsvision-embed-tessera-v2-nano, vision-embed-tessera-v2-small, vision-embed-tessera-v2-medium, vision-embed-tessera-v2-large, vision-embed-tessera-v2-teacher, vision-embed-olmoearth-v12-nano, vision-embed-olmoearth-v12-tiny, vision-embed-olmoearth-v12-small, vision-embed-olmoearth-v12-baseNo free-text taskGeospatial embedding inputs; track sensor, temporal, and preprocessing requirements.TESSERA and OlmoEarth embeddings guide (draft)Not verified
InsightFace Buffalo-Lvision-face-buffalo-lNo free-text taskFace analysis from imagery; input preparation and thresholds replace a prompting cookbook.InsightFace Buffalo-L guide (draft)Not verified
MoGe2, Video Depth Anything, and DA3vision-geometry-moge2-small, vision-depth-vda-small, vision-depth-vda-small-metric, vision-geometry-da3-smallNo free-text taskGeometry and depth estimation; track image and video preparation, scale, and camera conventions.MoGe2, Video Depth Anything, and DA3 guide (draft)Not verified
TripoSR, InstantMesh, and TRELLIS.2image-3d-triposr, image-3d-instantmesh-base, image-3d-trellis2-4bNo free-text taskManaged image-conditioned reconstruction; track source views, backgrounds, and masks.TripoSR, InstantMesh, and TRELLIS.2 guide (draft)Not verified

Music and audio

Table 6 lists provider sources and local recipes for music and audio models.

Table 6. Music and audio model guide coverage
Family and providerManaged model IDsSource status and linksScope and next actionLocal recipeChecked
ACE-Step 1.5 (ACE-Step)music-acestep, music-acestep-xl-base, music-acestep-xl-sft, music-acestep-xl-turbo, music-acestep-xl-turbo-lm4b, music-acestep-lm-1.7b, music-acestep-lm-4bGuide. Provider tutorialSeparate captions, lyrics, and metadata. Check XL base, fine-tuned, and turbo variants and auxiliary language model settings.ACE-Step 1.5 (ACE-Step) guide (draft)2026-09-04
MiniMax Music3 (MiniMax)music-minimax-music3Guide. Caption rewriter; model cardProvider prompt enhancement resource; map structured captions and separate lyrics to the local interface.MiniMax Music3 (MiniMax) guide (draft)2026-09-04
Magenta RealTime 2 (Google)music-magenta-rt2-small, music-magenta-rt2-baseReference. RealTime 2 model cardSource located. Extract style conditioning and live prompt-change examples.Magenta RealTime 2 (Google) guide (draft)2026-09-04
Muscriptormusic-muscriptor-small, music-muscriptor-medium, music-muscriptor-largeNo free-text taskMusic transcription; track audio preparation and output conventions.Muscriptor guide (draft)Not verified
RoFormer separation and cleanupmusic-separate-bs-roformer-viperx-1297, music-separate-bs-roformer-4stem, music-separate-mel-roformer-dereverb, music-separate-mel-roformer-denoiseNo free-text taskAudio separation, denoising, and dereverberation; track stem and model selection.RoFormer separation and cleanup guide (draft)Not verified
AP-BWE and UniverSRaudio-enhance-ap-bwe-16kto48k, audio-enhance-universr-audioNo free-text taskAudio enhancement; track sample rates and signal preparation.AP-BWE and UniverSR guide (draft)Not verified
Woosh generators (Sony AI)sfx-woosh-dflow, sfx-woosh-flow, sfx-woosh-vflow-8s, sfx-woosh-dvflow-8sReference. Woosh repositoryGeneration examples; separate text-only, video-conditioned, and duration-specific recipes.Woosh generators (Sony AI) guide (draft)2026-09-04
Woosh auxiliary modelssfx-woosh-clap, sfx-woosh-synchformerReference. Woosh repositoryScoring and conditioning components. associate text guidance with the consuming generator or CLAP query.Woosh auxiliary models guide (draft)2026-09-04
MMAudiosfx-mmaudio-large-44k-v2Reference. MMAudio repositoryExtract sound description and negative-prompt examples; verify text-only versus video conditioning.MMAudio guide (draft)2026-09-04

Video

Table 7 lists provider sources and local recipes for video models.

Table 7. Video model guide coverage
Family and providerManaged model IDsSource status and linksScope and next actionLocal recipeChecked
LTX 2, 2.3, and 2.5 (Lightricks)video-ltx-av, video-ltx23-av-mlx, video-ltx23-full-mlx, video-ltx23-a2vid-mlx, video-ltx25-distilled-bf16, video-ltx25-full-bf16Guide. LTX prompting guideShot, action, camera, and audio guidance; record version and hosted-only features before local adaptation.LTX 2, 2.3, and 2.5 (Lightricks) guide (draft)2026-09-04
Wan 2.2 (Wan)video-wan22-ti2v-5b-mlxReference. Wan2.2 repositoryTI2V-5B examples and prompt expansion; retain exact 5B controls.Wan 2.2 (Wan) guide (draft)2026-09-04
MiniMax H3 FL2VA and FastH3 (MiniMax)video-minimax-h3-fl2va-mlx, video-minimax-h3-fl2va-bf16-mlx, video-minimax-h3-fl2va-8bit-mlx, video-minimax-h3-fasth3-vsa-datafree-mlxGuide. Base video prompt guideExact H3 guide covers first-frame and last-frame alignment and audiovisual fields; review FastH3 settings separately.MiniMax H3 FL2VA and FastH3 (MiniMax) guide (draft)2026-09-04
MiniMax H3 Ref2VA (MiniMax)video-minimax-h3-ref2va-mlxGuide. Reference video prompt guideUse the reference-specific guide; validate reference ordering and identity descriptions locally.MiniMax H3 Ref2VA (MiniMax) guide (draft)2026-09-04
Cosmos3 Edge (NVIDIA)video-cosmos3-edge-mlxGuide. Cosmos3 prompt upsampling; Cosmos3 Edge model cardProvider JSON prompt preparation; separate image and video generation from reasoner and action tasks.Cosmos3 Edge (NVIDIA) guide (draft)2026-09-04
SCAIL-2 (Z.ai)video-scail2-14b-mlxReference. SCAIL-2 model cardSource located. Review reference image, motion, masks, and any prompt requirements.SCAIL-2 (Z.ai) guide (draft)2026-09-04
DreamX World (GD-ML)video-dreamx-world-5b-ar-mlxReference. DreamX World model cardSource located. Extract causal continuation and action-conditioning conventions.DreamX World (GD-ML) guide (draft)2026-09-04

Maintain the tracker

When a managed model or source changes, update the tracker in this order:

  1. Reconcile the exact IDs with the ManagedModelCatalog.allSpecs property and the canonical catalog.
  2. If a version needs a different prompt format, split its coverage into a separate row.
  3. From the original publisher's model card, review the linked prompting sources. Quantization mirrors establish artifact provenance unless their authors also publish independent prompting guidance.
  4. Record the source status, retrieval date, version scope, and next action. Identify failed retrievals and incomplete research explicitly.
  5. Update the local recipe with prompt structure, supported controls, sampling settings, examples, and known failure cases.
  6. In the Local recipe column, link the recipe with Drafted status.
  7. Validate the recipe with the exact managed model ID.
  8. Record the checkpoint revision, command, parameters, seed when applicable, and reviewed output.
  9. Link the validation evidence in the Local recipe column.
  10. Change the recipe status to Validated.

Start recipe validation with SenseNova U1.5, the two H3 guides, the Liquid AI text and vision guides, ACE-Step, MiniMax Music3, and Cosmos3 prompt preparation. These sources describe model-family conventions. Before using Krea or LTX hosted controls, verify that the corresponding local control exists.

Released under the MIT License.