SyQon Studio and DeepPrism V2 (OSC)

Rig context: ASI585MC Air on the FRA400 or 107PHQ via ASIAIR, plus the Seestar S30 Pro. SyQon Studio 1.1 public beta, DeepPrism V2, Parallax and Axiom V3, with PixInsight 1.9.x as the alternative host.

The workflow at a glance

Stage 0: Set up

1. Install Studio and sign in. Free account; let Required Items download models and catalogues.
2. Add the PixInsight script. SyQon-Hub repository, then Script > SyQon > Studio.

Stage 1: Stack

3. Build the session. Classic, filter RGB, one acquisition per night or rig, scientific defaults.
4. Analyse and review subframes. Slow step; sort by quality, check the tails, confirm Use.
5. Analyse and Stack, drizzle Off. Then read coverage, rejection map and manifest.

Stage 2: Flatten and calibrate

6. Remove the gradient. Deep Gradient or Gradient Removal, to a new layer.
7. Image Solver, then SPCC. Local ASTAP solve, Gaia XP colour calibration.

Stage 3: Restore linear

8. Deblur with Parallax. Deblur strength 0.45 to 0.5; or BlurXTerminator in PixInsight.
9. DeepPrism Advanced. Linear only; careful, keeps faint detail.

Stage 4: Stretch and clean

10. Stretch. Continuum Stretch in Studio, or my usual PixInsight stretch.
11. DeepPrism Ultra. Stronger pass now faint signal is lifted; new layer.

Stage 5: Judge and save

12. Compare, then save the project. Fit and 100%; SYQ or SYQPRJ before any export.

Walking noise: DeepPrism Ultra removes it on linear or stretched data. Max only when time allows.

No DeepPrism licence: Essential (free) does the linear pass. Axiom Mini and Parallax Nano are also free.

The logic in one line

Flatten and calibrate colour before any neural model touches the data, deblur before denoise, and split denoising so the gentle linear pass protects faint signal before the strong pass cleans the stretched image.

Non-negotiable

• Essential and Advanced need linear input. Never run them after the stretch.
• Gradient removal before SPCC, and a plate solve before SPCC.
• Deblur before any denoise, never after.
• Drizzle Off for the first integration. Turn it on only for real, well spread dithers.
• Keep darks, bias and flats inside their own acquisition. Never borrow across nights.
• Every neural step to a new layer, checked at 100%.

Your call

• Host: Studio end to end, or stack elsewhere and call the models from PixInsight.
• Max versus Ultra: Max is far slower for a very small gain on normal data.
• Deep Gradient versus classical Gradient Removal around large targets.
• Prism application: 1.0 for full effect, lower to keep some grain.
• Star separation with Axiom before or after the second Prism pass.
• Drizzle 2x on dithered Seestar data; off on the FRA400 and 107PHQ.