SFUMIA / Documentation
Guide
Prepare an AI generation
Install a local model or make an informed choice about a Cloud job.
In this guide
Install models in settings
Compose your workflow, then open the AI installation panel in SFUMIA settings. Check the complete profile size, dependencies, hardware requirements and official model terms.
Explicitly install the components you want. Some hosts require you to accept terms on their own website. SFUMIA does not accept them for you or bypass denied access.
Choose a local engine
Native local nodes run directly in SFUMIA without requiring ComfyUI. The dedicated ComfyUI node remains an optional integration for workflows that use it.
Check the profile status before running a job. Technical preparation or a trial on different hardware does not guarantee generation on your PC.
What the local catalogue offers
Model-based local nodes appear under « IA → IA Image (Local) », « IA Vidéo (Local) » or « IA 3D (Local) ». A node whose engine cannot be installed yet shows « Ce node peut être ajouté, relié et enregistré. Son moteur local n’a pas encore d’installation prise en charge dans cette version. »
| Node | Status in this version |
|---|---|
| « IA FLUX Image » | FLUX.2 Klein in Local mode: engine targeted by this private package; validated on a single PC. |
| « IA Qwen Edit » | Local image editing, still under qualification; Local mode is reserved for Labs access. |
| « IA SAM Segmentation » | Local segmentation masks; model to install from settings. |
| « LDR → EXR · Pro » | Dynamic range expansion to a 32-bit EXR; requires the advanced EXR permission. |
| « IA Upscale », « IA Inpaint », « IA Outpaint », « IA Carte de profondeur · Labs », « IA Colorisation · Labs » | Present in the catalogue; their local engine has no supported installation in this version yet. |
| « IA Wan · Outils natifs », « IA LTX · Local », « IA MiniMax · Local » | Local video and sound still under qualification; no supported installation in this version. |
Choose a Cloud job
A Cloud job requires the selected product to be available and authorised for your account. A name in a menu does not prove the service is open; read its status and quote before confirming.
Review the media being sent, the quoted cost and the budget in use before confirming. A Cloud job must not silently replace local generation.
Studio permissions and allocations are separate from your personal budget. Keep the current result, then decide separately whether to version or publish it.
In this version, SFUMIA Cloud is still under qualification. The interface shows « Les tarifs SFUMIA Cloud sont en cours de vérification. » or « Cette fonction SFUMIA Cloud est en cours de qualification. », and the cost appears as « Prix indisponible ». Until prices are reviewed, no quote is served and no Cloud job is started. Prices may change according to the cost of our AI model providers.
Some nodes accept the API key of the model publisher, for example the « API Google (BYOK) » mode of the « IA Nano Banana » node: the computation is then billed by that publisher, outside SFUMIA Cloud, and your media are sent to it.
Cloud audio: voice, dialogue, music and sound effects
The « Voix », « Dialogue », « Musique » and « Bruitages » nodes in the Sound category list Cloud models: Eleven v3, Eleven Multilingual v2 and Turbo v2.5, MiniMax Speech 2.8, Inworld TTS, Kokoro, Eleven v3 Dialogue, Eleven Music, MiniMax Music 2.6 and 2, Stable Audio 3 Small and 2.5, Lyria 2, Eleven Sound Effects v2 and Stable Audio 3 Small SFX. The catalogue is visible without credits; Cloud execution remains under qualification and no quote is served until prices are reviewed.
Limits shown by the interface
« Dialogue multi-voix » accepts 32 lines and 5,000 characters for the dialogue, with « Ajouter une réplique ». Eleven Music produces 3 to 600 s and every minute started is billed; Lyria 2 produces 30 s of instrumental music; Stable Audio 3 Small 1 to 120 s; Stable Audio 2.5 1 to 190 s; Eleven Sound Effects v2 0.5 to 22 s; MiniMax Music chooses the duration. Prices may change according to the cost of our AI model providers.
Available features depend on your version and installation.