HISTORICAL DIG · JUL—AUG 2026
What should a first local AI image prove?
“Local image generation” sounds like one switch. In practice it spans app installation, model download, compute location, output storage, device time and heat, optional cloud routes, and accounts. Judging only the first image's appearance hides those decisions.
HOW IT RELATES TO WORKFLOW 001
Relation to the current subject
It first asks whether the compute path is controllable—not whether the work can be published.
This historical record shares one principle with Workflow 001: choose a path by the task, not by feature count. Its direct link is the runtime-location question. A model download may use the network while the generation run itself is explicitly local.
Record state at the time
- Creation & local AI
- Local visual workflow
- Standard record
- Developing
- Historical E2 · earlier cross-source label · no current install, model download, or generation
THE GUIDANCE AT THE TIME
Judgment at the time
The first image should prove the path, not your skill.
Open full rationale and return rules
The first image should prove the path, not your skill. Use a text-only prompt with no people, trademarks, or client information, choose local compute explicitly, and record where the model came from, how long the run took, what happened to the device and storage, where the file landed, and how to delete it. Visual polish comes later.
- Who it was for
- Creators with an Apple device who want to judge whether local image generation fits their workflow before importing people, brand, or client material
- Why it mattered then
Open note
The first trial is valuable not because it creates a publishable work, but because it exposes download, compute, storage, device load, and exit cost with little risk. If those remain unclear, importing real material only makes rework more expensive.
- Where the judgment stops
Open note
BitShovel has not installed Draw Things, downloaded a model, generated an image, or tested speed, traffic, heat, battery, saving, or deletion. The maker now separates local compute, free Cloud Compute, and paid Draw Things+ routes; free local generation can still begin with a model download from its servers. “Local compute for this run” does not mean “zero network for the app's entire lifecycle,” nor does it establish privacy, security, performance, or publication rights. The earlier publication, brand-display, and removal-and-rework cases answer a release-workflow question, so Revision 5 removes them from this first-run decision.
CHECK BEFORE READING ON
Fit check
Continue only when the low-risk sample, compute mode, and exit cost are all understandable.
This record may help
- The device has enough free storage and time for a model download followed by one disposable sample.
- The first trial can stay text-only without granting photo-library, contacts, cloud, or client-material access.
Do not use it yet when
- The first step already requires a real person's photo, brand asset, unpublished design, or client file.
- The device has no comfortable storage, memory, temperature, or battery margin and cannot absorb a large model download.
- You need this single trial to establish commercial use, publication, delivery, or reproduction of maker-selected sample quality.
HISTORICAL TRIAL · UNRUN AND UNFROZEN
Historical trial
Use one text-only prompt to complete a local path that can be stopped and deleted.
- Time needed
- 20–40 minutes plus model-download time (earlier estimate)
- Likely cost
- Choose local compute first; buy no subscription or cloud credits
- Permission boundary
- No sign-in, Cloud Compute, people, trademarks, client material, or photo-library import
- 01
First step
Open this step
Choose a model the device can comfortably handle and an explicit local-compute route. Generate one low-risk sample from a text prompt describing only an abstract object and lighting, while recording the model source, size, compute mode, and start and finish time.
- 02
What should exist
Open this step
One disposable sample plus a short note covering compute mode, download, elapsed time, storage change, device condition, project location, and output location.
- 03
How to tell it worked
Open this step
You can reopen the project and output, point to local rather than cloud compute for this run, and explain how to stop, delete the sample, and handle an unneeded model. Network notes report only observed requests and never claim absolute zero network.
- 04
Stop when
Open this step
Stop if heat, memory, or storage pressure becomes uncomfortable; the app asks for sign-in or cloud credits; compute mode is unclear; or the trial starts requesting real material.
- 05
How to back out
Open this step
Stop generation, delete the sample and test project, decide whether the model is still needed, revoke unnecessary photo access, and record why the trial did not finish.
BITSHOVEL R5 EXPLAINER · NOT A PRODUCT INTERFACE
Method and interfaces
Put model download, local compute, the cloud fork, and the real-material gate in one view.
Open method boundary
The citrine path keeps model download visible as a possible network step. The red fork requires an explicit local-versus-cloud choice for this run. People, brands, client material, and publication rights stay beyond the next gate.
The citrine connection keeps model download visible as a network step; the red fork separates local compute from Cloud Compute. The sample stays text-only, while people, brands, client material, and publication rights remain outside the next gate.
About this image and its source
BitShovel R5 workflow explainer created August 15, 2026 from current Draw Things pricing, privacy, App Store, and model documentation. It is not a Draw Things interface, BitShovel run, performance test, privacy audit, or endorsement.
Open source page ↗SOURCES AND EVIDENCE
Sources and evidence
As of August 15, 2026, five current maker and platform sources only define local, download, cloud, account, and storage routes. There is no BitShovel run or independent performance result.
Open evidence boundary
The historical E2 came from maker material read alongside three publisher outcomes. Revision 5 no longer piles those release cases into a first local trial. The current page uses maker and platform material only to define local, download, cloud, account, and storage routes; it has no BitShovel run or independent performance result.
- 01Current product tiers · Rechecked
Draw Things pricing: local compute, free cloud, and paid cloud are separate routes
Open source boundary
The maker says the free local edition downloads models from its servers and generates on-device; Community can offload work to managed servers, while Draw Things+ adds cloud limits and LoRA upload.
Open original source ↗ - 02Current privacy statement · Rechecked
Draw Things privacy: on-device data, optional account, cloud LoRAs, and subscriptions are handled separately
Open source boundary
This is a maker statement, not an independent audit. It also confirms that Cloud Compute can involve account identifiers, cloud storage, and subscription records.
Open original source ↗ - 03Current platform listing · Rechecked
App Store: local offline capability and optional Cloud Compute coexist
Open source boundary
The platform listing carries the developer's local-generation and optional-cloud descriptions and notes that privacy information is developer-provided rather than verified by Apple. It is not a BitShovel device test.
Open original source ↗ - 04Current model documentation · Rechecked
Model files use material storage and have separate download, import, and deletion locations
Open source boundary
Project documentation describes models ranging from hundreds of megabytes to nearly 10GB and separates downloads, imported models, and project history. Exact versions and devices still need an in-app check.
Open original source ↗ - 05Current release record · Rechecked
Current releases continue evolving both local models and Cloud Compute
Open source boundary
The release page shows continued changes to local models, device optimization, accounts, remote providers, Cloud Compute, and tier names, so compute mode must be rechecked in the current interface before a trial.
Open original source ↗
REVISION RECORD
Revision record
Revision 5 separates the first local trial from publication rights.
Open 4 revisions
- Revision 2
Changed a product recommendation into one low-risk first test focused on local use.
- Revision 3
Added publication, brand-display, and removal-and-rework outcomes to connect technical paths with pre-publication rights checks.
- Revision 4
Narrowed the title to the first local-image decision while preserving the uninstalled, ungenerated, maker-media boundaries.
- Revision 5
Separated local generation from an app-wide zero-network claim, moved publication and brand cases to the release-rights topic, removed three product images without open reuse rights, and used a BitShovel compute-path explainer instead.
- What the record said before
- A first local image should also explain real publication, brand display, removal and rework, data routes, and publication rights.
- What changed
- The first trial answers only whether the compute path is controllable: how the model reaches the device, where this run computes, what the device and storage spend, where project and output live, and how to stop and delete. Publication rights need a separate Dig.
WHAT THIS RECORD LEAVES
What this record leaves
Prove the path before bringing real material into it.
If the first run cannot make compute mode, model size, project location, and deletion clear, a polished image is not enough to support adoption. People, brands, client material, and commercial publication need a separate rights and disclosure check.
Back to the Dig index