Veles

Veles

AI Image Generation Platform

Veles Veles
V — / 2000
Edit Preset
Generation Parameters
Video Parameters
LoRA Strengths
Refiner

How to generate

1. Choose a PresetPick one that matches the kind of image you want.
2. Write your promptDescribe what you want to see in plain words.
3. Adjust settingsOnly change them if you know what you are doing.
4. Click GenerateThe network finds a matching worker automatically.
5. Download the resultSave it or generate again with a different prompt.

Base Model Guide

FLUX-S FLUX.1 Schnell
Distilled diffusion transformer. Excellent at rendering readable text and typography, bold color palettes, and strong prompt adherence.
✓ Text-in-image, logos, posters · vivid output · Schnell runs in 4 steps
✗ No negative prompt (zeroed conditioning) · can look over-stylized on faces
SDXL Stable Diffusion XL 1.0
The workhorse: huge LoRA ecosystem, versatile across realism, anime, and concept art. Optional refiner pass adds polish.
✓ Largest community LoRAs · broad style coverage · fast on consumer GPUs
✗ Unreliable text rendering · weaker prompt adherence than newer models
SD 3.5 Stable Diffusion 3.5 Medium
MMDiT architecture with improved prompt fidelity and composition vs SDXL. Permissive community license.
✓ Better prompt following · good typography · lightweight MMDiT
✗ Small LoRA ecosystem · prefers low CFG (≈ 4.5) + sgm_uniform scheduler
Job Summary
Price —
Max VRAM —
Gen Time —
Execution time only · No queue wait included
Weekly allowance —
Remaining after —
Resets in —

Live Network

Jobs in Queue—
Online Workers—
Avg Queue Wait—
Avg Job Time—
Start with a preset
Choose what you want to create.
No preview media available for this preset yet.
Community Presets
Create a Preset →
Select a Preset above to start generating

Source Image

Mask Source
Drop mask image here or click to upload
Uploaded mask preview
mask
Mask preview (white = will be regenerated)
mask preview
SAM Detection Settings
AI detection: mask generated from Detection Prompt using SAM + GroundingDINO — no painting needed.
▸ Advanced Mask Controls
Outpaint Padding (leave 0 for pure inpaint)

Prompt

Trigger words:

Shared controls

Generation price
—
Select a preset to see price
Advanced Controls expand

Model Slots

Parameters

More steps = higher price

Video Settings

LoRA Strengths

Refiner

Import / Export Settings

Recent Jobs

No jobs yet

How models work

1. Submit for reviewOpen Submissions and give us the model page URL and a direct file URL.
2. Automatic license checkWe check the upstream license and Veles-relevant rights before anything is queued.
3. Admin reviewOnce automatic checks pass, an admin approves it. Non-commercial models are blocked.
4. Reuse everywhereApproved models appear in My Models and can be used in any Preset. Workers download them automatically.

Browse Models

Models are filtered by Base Model when selecting in templates. SDXL 1.0 templates show only SDXL models; WAN Video templates show only WAN models. Models with no Base Model set are shown in all templates.

Blocked Models (license / policy)

These models are blocked from installation and use on the network for licensing or policy reasons. You can still review source information on CivitAI.
Loading blocked list…

Want to propose a new model?

Models are no longer self-registered. Submit the model page URL and a direct file URL; we verify the upstream license, then an admin approves or rejects it. Non-commercial licences are blocked.

My Models

Your registered models — manage versions, edit metadata, or delete.

Filters & search

Loading…

How templates work

1. A template = workflow blueprintIt defines how generation is wired together internally.
2. Named model slotsEach slot is a place for a specific model (checkpoint, LoRA, VAE…).
3. Templates are reusableMany Presets can share the same template.
4. Use a templateClick Use to go to Generate with that template pre-selected.

Workflow Templates

Reusable generation blueprints used by Presets. Each template defines the model slots, task type, and internal workflow.

Currently active: Supported base models — SDXL 1.0, FLUX.1 Schnell, WAN Video 2.2 TI2V-5B, and SD 3.5 Medium. Other families (SD 1.5, Pony) are not wired into the current generation workflows.

How Presets work

1. Pick a templateChoose the workflow that matches what you want to generate.
2. Attach modelsFill each slot with the right model from your library.
3. Set defaultsAdd a default prompt and generation parameters.
4. Upload previewsGood preview images build trust and show what the Preset can do.
5. Use itClick Use to apply this Preset to a new generation.
Only you can edit or delete your own Presets.
No Presets yet — create one below

Filter presets

Presets combine a workflow template, selected models, and default generation settings

Create Preset

Generation Parameters defaults applied when this preset is used
Video Parameters
LoRA Strengths
Refiner
Save the preset first, then add preview images.

Preview Images

How history works

1. Review resultsEvery generation you run is saved here automatically.
2. Download againClick any result to download or preview it again.
3. Check failuresIf a job failed, open its details to see the error.
4. Compare outputsUse this page to compare runs and find what works best.

Generation History

Loading…

Your account

ProfileYour email and sign-in details.
PresetsManage the Presets you have created.
PreferencesSet auto-download, auto-open, and other convenience options.
Settings are saved in your browser.

Profile

Veles

Veles

✓ Signed in

Submit an asset

1. Direct file URLA direct link to the .safetensors / .ckpt download (HuggingFace, CivitAI, etc.).
2. Model page URLThe page describing the model — license, previews, author.
3. We preflight itLicence checks run automatically. If we can't approve, we'll tell you why.
4. Admin reviewAn admin will approve, reject, or request more info.

Submit a new asset

Advanced types (text encoders, diffusion models, CLIP/UNet) — pick Other and note the exact type; admin review will route it.

My submissions

Request support

What is this?Public list of models that people have submitted but Veles does not yet support. Weight tags, sizes, or runtimes that we haven't built in yet end up here.
Why vote?Your vote signals which unsupported model/variant should get attention first — engineering effort and licence negotiation both follow demand.
What blocks a model?Each entry shows whether the blocker is a missing weight/quant runtime, a missing workflow, or a licence clarification / negotiation.

Unsupported model ranking

Admin review

ApproveRe-runs preflight, creates an asset + version row.
RejectRejection is terminal. Requires a reason code + note.
Request infoPause the submission and ask the submitter for clarification.
BlockedPreflight auto-blocked at submission time. Cannot approve until the source is compliant.

Submission queue

Select Model

Model
Loading…
Add Version —
LoRA Defaults
Confirm

Paint Mask
White = area to generate · Black = keep