Spark Fuse™
Overview
Spark Fuse™ is instant GPU and CPU serverless compute for creative production, built natively into Spark Cloud Studio.
Use it when a job does not fit a traditional render farm queue: LoRA training, ComfyUI batches, simulations, neural textures, SAM2 roto, upscaling, inference across thousands of frames, and other short-to-medium GPU or CPU workloads. Submit a job, watch live logs, and find finished outputs in ShareSync™.
Fuse is purpose-built for creative pipelines: DCC-aware workflows, connected secure storage, and the same login and billing as the rest of Spark.
Key Features
Instant serverless
compute
Run GPU and CPU jobs without managing machines or queues.
- Launches in seconds on InstantCompute
- Scales for iterative and batch workloads
- Pay only while your job runs
Native creative
pipeline
Built for production workflows, not generic cloud VMs.
- ComfyUI integration and DCC-friendly jobs
- Outputs land directly in ShareSync™
- Live log streaming while jobs run
Two compute
engines
Pick the engine that matches speed vs cost for each job.
- InstantCompute for rapid iteration and debugging
- SmartCompute for larger, interrupt-tolerant batches
- GPU and CPU configurations across regions
Unified Spark
ecosystem
Same identity, billing, and storage as the rest of Spark.
- One Spark Cloud Studio account and billing
- ShareSync-backed inputs and outputs
- API-driven job submit, monitor, and cancel
Compute modes
Spark Fuse offers two ways to run jobs:
| Mode | Best for | Typical launch | Pricing (from) |
|---|---|---|---|
| InstantCompute | Rapid iteration, debugging, inference tests | ~1-2 seconds | GPU from $0.49/hr · CPU from $0.19/hr |
| SmartCompute | Bursting, interrupt-tolerant, batch processing | A few minutes | GPU from $0.29/hr · CPU from $0.09/hr |
Rates vary by GPU/CPU type, region, demand, and availability. Detailed estimates are available through the Spark API and on the Spark Fuse product page.
How It Works
- Submit: Send a job (image + command + instance type) via the Spark Fuse API or an integrated tool such as the ComfyUI node.
- Prepare & run: Fuse selects compute, starts the workload, and streams live logs.
- Write outputs: Everything written to the job output path is persisted on ShareSync.
- Review & iterate: Pull results into your workstation or pipeline, then submit the next job.
Use Case Scenarios
- Compositing: Offload comps, resizes, denoise scripts, matte generation, and automated Nuke workflows.
- AI / Lookdev: Train and compare LoRA variants; run ComfyUI workflows as batches with organised ShareSync outputs.
- Simulations: Fire debug wedges on InstantCompute, then send full batches to SmartCompute once settings are dialed in.
- ML & neural work: Inference at scale, Gaussian splats, neural texture baking, SAM2 roto, overnight upscales.
Getting started
- Use Fuse from your pipeline: Using Spark Fuse through the API
- Product overview and guide pricing: sparkcloud.studio/fuse
- ComfyUI integration: spark-fuse-comfyui-node
- Access and onboarding: Contact Spark
Spark Fuse is available through early access. Contact Spark Cloud Studio to get started.
Summary
Spark Fuse™ fills the gap between a workstation and a render farm: serverless GPU/CPU compute that starts in seconds, streams logs live, and writes results straight into ShareSync, so creative teams can iterate faster without stitching compute, storage, and delivery together by hand.