Browse by industry, then open a project brief pre-filled for your scenario.
SaaSLanguage & vision-language
Enterprise knowledge assistant
Grounded answers from wikis, SOPs, and tickets.
We fine-tune an open LLM for your terminology, answer format, and citation behavior, and pair it with a tuned embedding model for retrieval. Knowledge stays in your index rather than in the weights, so it can be updated without retraining. Evaluation sets come from real employee questions.
Models: Llama 3.x, Qwen3, or Mistral, with BGE or E5 embeddings
Deliverables: Fine-tuned weights, retrieval evaluation report, deployment guide
Start this use caseSaaSLanguage & vision-language
Customer support copilot
On-brand reply drafts with clear escalation rules.
Trained on historical tickets and macros to draft replies in your tone, across languages. Preference tuning reduces off-brand phrasing before an agent reviews each draft.
Models: Gemma 3, Llama 3.x Instruct, Phi-4-mini
Deliverables: Fine-tuned checkpoint, tone rubric, API integration notes
Start this use caseFinanceLanguage & vision-language
Contract and compliance review
Clause extraction for attorney review.
Long-context fine-tunes that flag risky clauses and summarize agreements. Output is structured for attorney review and logged for audit. The model supports counsel and does not replace them.
Models: Llama 3.x (70B class), Qwen long-context models
Deliverables: Domain adapter, structured-output prompts, red-team report
Start this use caseHealthcareLanguage & vision-language
Healthcare administration
Note summarization on de-identified data.
Administrative support only: note summarization and coding suggestions on de-identified data, with clinician review. We do not build diagnostic or treatment models.
Models: Open bases selected after a data-use review
Deliverables: Training plan for de-identified data, evaluation on approved samples
Start this use caseFinanceLanguage & vision-language
Financial research and reporting
Earnings and portfolio Q&A on licensed data.
Models tuned on your research corpus, evaluated for refusal behavior and data freshness. Reasoning-oriented distills support multi-step report drafting, with an analyst reviewing every output.
Models: DeepSeek-R1 distills, Qwen3, Mistral
Deliverables: Checkpoints, benchmark against an analyst-written reference set
Start this use caseSaaSLanguage & vision-language
Code assistant for private repositories
Completion that knows your internal APIs.
Fine-tuned on your internal libraries and style guides. We scan for secrets and train only on repositories you approve.
Models: StarCoder2, DeepSeek-Coder, Code Llama
Deliverables: Code model, private evaluation harness, IDE integration notes
Start this use caseSaaSLanguage & vision-language
Sales enablement and proposals
Battlecards and RFP drafts from CRM exports.
Preference tuning with DPO aligns drafts with your brand voice and approved compliance language. Sales teams receive drafts inside existing tools; nothing is sent automatically.
Models: Mid-size instruct models with preference data
Deliverables: Aligned checkpoint, prompt library, CRM field-mapping document
Start this use caseManufacturingLanguage & vision-language
Visual inspection reports
Defect notes from production-line images.
Vision-language models that return structured JSON for MES integration. Training uses your labeled defect library, with class-balance reporting.
Models: Qwen2.5-VL, InternVL2
Deliverables: Fine-tuned vision-language model, JSON schema, offline evaluation on held-out lines
Start this use caseRetailImage & video
E-commerce product imagery
Consistent lighting and backgrounds across a catalog.
LoRAs trained on your packshots keep backgrounds and lighting consistent at catalog scale. ControlNet preserves product shape and proportions.
Models: SDXL or FLUX.1 LoRA with ControlNet
Deliverables: LoRA pack, inference presets, brand-safety filter
Start this use caseRetailImage & video
Fashion and apparel lookbooks
A campaign look captured in a LoRA.
Separate LoRAs for fabric, lighting, and model identity, combined where licensing and consent allow. Creative teams receive a scored style matrix.
Models: SDXL or FLUX.1, multi-LoRA
Deliverables: Style LoRAs, prompt guide, review grid
Start this use caseReal estateImage & video
Virtual staging and interiors
Furnished rooms with the layout left intact.
Depth- and edge-conditioned generation furnishes empty rooms without moving walls or openings. Optional on-image disclosure supports listing-service rules on virtually staged photos.
Models: SDXL with ControlNet (depth, canny)
Deliverables: LoRA and ControlNet bundle, staging workflow guide
Start this use caseMediaImage & video
Concept art for games and entertainment
Character and environment style packs.
Trained on your style bible and approved asset libraries. Fixed-seed workflows support controlled, repeatable iteration.
Models: SDXL, FLUX.1, Stable Diffusion 3.5
Deliverables: Style LoRAs, generation guide, content-policy filter
Start this use caseMediaImage & video
Marketing creative at scale
Ad variants within brand guidelines.
FLUX.1 [schnell], which is Apache 2.0, handles volume work. FLUX.1 [dev] is used for hero assets only where a commercial license from Black Forest Labs is in place. Review rubrics score color adherence and logo placement.
Models: FLUX.1 [schnell] and [dev]
Deliverables: Two-tier workflow, brand-check prompts
Start this use caseReal estateImage & video
Architectural visualization
Facade and landscape variations.
ControlNet holds structural lines while you explore materials and surroundings. Useful for client presentations and early design reviews.
Models: SDXL with ControlNet
Deliverables: Material LoRAs, control presets
Start this use caseMediaImage & video
Personalized characters and avatars
Identity-consistent generation, with consent.
DreamBooth-style training only with documented consent and likeness rights. We decline projects that involve a person's likeness without it.
Models: DreamBooth or FLUX.1 character LoRA
Deliverables: Character adapter, consent checklist, usage policy
Start this use caseRetailImage & video
Packaging and print mockups
Fast visual drafts for review.
IP-Adapter and reference conditioning keep a consistent brand look across substrates and finishes. Diffusion models cannot reliably reproduce exact logos or label text, so final artwork is composited in prepress.
Models: SDXL with IP-Adapter
Deliverables: Mockup LoRA, reference-image pipeline
Start this use caseMediaImage & video
Short-form video loops
Product spins and motion posters.
AnimateDiff and CogVideoX fine-tuned on short clips from your brand library. Open video models generate a few seconds at a time, so we cover upscaling and frame interpolation for platform formats.
Models: AnimateDiff, CogVideoX
Deliverables: Video LoRA, render recipe, export settings
Start this use caseCross-industryInfrastructure
Private-cloud and on-premises training
Your data stays in your environment.
We train inside your AWS account or Modal workspace using credentials you control, and artifacts are written to your own storage. Suited to teams with data-residency or regulatory requirements.
Models: Open model of your choice
Deliverables: Runbook, IAM template, trained weights in your storage
Start this use caseCross-industryInfrastructure
Adapter libraries and merges
Multiple LoRAs, tested together.
A catalog of adapters with compatibility notes and tested merged presets, so creative and ML teams can combine styles predictably.
Models: SDXL and FLUX.1 ecosystems
Deliverables: LoRA library, merge recipes, evaluation sheet
Start this use case