LLM Assistance for Imaging

FILM provides machine-readable reference files that describe the hardware on our microscopes — every objective, laser line, detector, filter, and software module. Upload a reference file into a large language model (LLM) and ask it to help you plan your experiment. The LLM will cross-check your sample, fluorophores, and acquisition plan against the actual components on the microscope and flag any issues before you sit down at the system.
This works for everyone: MSc students planning a first confocal Z-stack, PhD students designing a tiling experiment, and postdocs setting up FLIM or multiphoton workflows.
It is not a replacement for hands-on training or FILM staff support. It is a planning tool that helps you arrive at your booking better prepared.
Available system references
System
File
Modalities
HCF4 — Nikon AX R MP NSPARC
Confocal, resonant, multiphoton, FLIM, FRET, FCS, SHG, ISM super-resolution, widefield, DIC
HCF2 - Leica Stellaris 8
Confocal, resonant
We are building reference files for the remaining FILM microscopes (HCF2, HCF3, CF6, HSR1, WF5, WF6, LS1, HMI1, and others) and for analysis software (Fiji, Imaris, Huygens, CellProfiler). These will appear here as they become available.
How to use it
  1. Download HCF4.md from the link above.
  2. Open claude.ai (free tier available) or another LLM of your choice.
  3. Upload the file into the conversation, or add it to a Claude Project so it persists across sessions.
  4. Describe your experiment — your sample, fluorophores, and what you want to measure — and ask your question.
Example prompts by experience level
Everyday confocal and widefield — MSc / new users
"I need to image my fixed tissue section — where do I start?"
I have a 10 µm cryosection of mouse lung on a glass slide, stained with DAPI and Alexa Fluor 488 phalloidin. I just want a nice overview image of the whole section. What laser lines, filters, and objectives should I use?
The LLM will identify the 405 nm and 488 nm lines from the LUA-S6 laser unit, recommend the descanned detector with appropriate spectral windows, and suggest the 10× or 20× air objective for an overview. It will also explain that you can tile the section using the motorised stage.
"How do I tile a large tissue area?"
My tissue section is about 8 mm × 5 mm. I want to acquire it at 20× with a Z-stack of 10 slices at 1 µm spacing, then stitch it into a single image. How long will this take?
The LLM will estimate the number of tiles based on the 20× field of view, factor in overlap for stitching, multiply by the Z-stack, and give a rough acquisition time for galvano vs resonant scanning.
"Which objective do I pick for my coverslip?"
I have cells on a #1.5 coverslip in a 24-well plate. Do I use the 60× oil or the 40× water immersion for fixed cells stained with three colours?
The LLM will compare the two objectives by NA, working distance, and resolution, and explain which suits fixed cells on a standard coverslip.
"Can I see my three labels without bleed-through?"
My cells are stained with DAPI, Alexa Fluor 568, and Alexa Fluor 647. Will the channels bleed into each other on the confocal?
The LLM will map each dye to a laser line, suggest sequential vs simultaneous acquisition, and recommend spectral windows that separate the three emission peaks.
"How thick should my Z-steps be?"
I want a confocal Z-stack of GFP-expressing cells at 60× oil for a 3D reconstruction. What Z-step size should I use?
The LLM will calculate the Nyquist sampling interval based on objective NA, emission wavelength, and pinhole setting.
"How do I set up DIC?"
I need to check cell morphology using DIC before fluorescence. Which DIC components are on HCF4 and which do I use for each objective?
The LLM will list the installed DIC sliders for each magnification and explain the required polariser and analyser combination.
"I want to image live cells — what do I need to turn on?"
I want a 4-hour time-lapse of GFP-expressing HeLa cells. What keeps them alive on the microscope and how do I stop focus from drifting?
The LLM will recommend the cage incubator, Perfect Focus System, and if relevant the water immersion dispenser, and suggest the resonant scanner for reduced phototoxicity.
Standard confocal and multiphoton — MSc / PhD
"Can I do this four-colour protocol?"
I want to image DAPI, AF488, AF568, and AF647 in cleared mouse brain at 25× silicone immersion, Z-stack through 500 µm. Can HCF4 do this?
The LLM will check laser lines, confirm the silicone objective is installed, verify four-channel separation on the descanned detector, and estimate acquisition time.
"Is my tiling plan realistic for my booking?"
I want a 6 × 6 tiled Z-stack (30 slices, 2 µm spacing) of mouse kidney at 20×, two channels, galvano 1024 × 1024. How long and is there a faster way?
The LLM will estimate total time and suggest whether resonant scanning or lower resolution would be acceptable.
"Help me plan a multipoint time-lapse"
I have a 6-well plate, 5 positions per well, every 30 minutes for 12 hours, Z-stack of 5 slices, GFP only. Can I fit this into each interval? What JOBS setup do I need?
The LLM will calculate whether 30 positions fit within the 30-minute window, recommend resonant scanning for speed, and outline a JOBS workflow.
Advanced FLIM and multiphoton — PhD / Postdoc
"Can I do two-photon FLIM of metabolic cofactors?"
I want two-photon FLIM of NAD(P)H (750 nm excitation, 450/70 emission) and FAD (890 nm, 550/88 emission) in live cells. Can HCF4 do this?
The LLM will verify the laser tuning range covers both wavelengths, confirm both NDD filter cubes are installed, and check PicoQuant FLIM detector availability.
"Which detector path for FRET-FLIM?"
I'm using mTurquoise2 (donor) and mVenus (acceptor). NDD or descanned for the cleanest donor lifetime?
The LLM will compare NDD collection efficiency against descanned spectral flexibility and recommend the best path for your fluorophore pair.
"Is my FLIM acquisition plan optimal?"
512 × 512, 8 µs pixel dwell, 60 frame accumulation, 4 × 4 tile, 10 % overlap, every 15 minutes for 6 hours. Does this make sense?
The LLM will estimate total frame time, flag scanner compatibility with FLIM, check PFS viability over 6 hours, and suggest JOBS automation.
"Help me build a FLIM screening workflow"
Screen a 96-well plate for metabolic FLIM signatures: autofocus each well, single-plane two-photon FLIM at 750 nm, review lifetime maps afterwards. What JOBS nodes?
The LLM will outline the full JOBS Editor workflow from well-plate definition through to JOBS Viewer review.
"Confocal FLIM vs two-photon FLIM?"
For GFP lifetime in thick tissue — confocal at 488 nm with pinhole, or two-photon at 920 nm with NDD? Pros and cons on HCF4?
The LLM will compare depth penetration, scattering, photobleaching, detector sensitivity, and laser power at each wavelength.
"Can I do SHG alongside FLIM?"
I want collagen SHG simultaneously with NAD(P)H FLIM in a skin biopsy at 750 nm excitation. Which detector picks up the SHG?
The LLM will identify the SHG signal at 375 nm, check whether the diascopic detector can collect it, and confirm the NDD handles the NAD(P)H channel.
Tips for getting useful answers
Don't be afraid to ask basic questions. "What does the pinhole do?" or "Why sequential scanning?" are good prompts. The LLM will explain using the actual HCF4 hardware rather than giving a generic textbook answer.
Be specific about your sample. "Live cells" is too vague. "Live primary mouse macrophages on a 35 mm glass-bottom dish, labelled with mClover3-tagged protein X" lets the LLM reason about coverslip compatibility, immersion medium, and objectives. For fixed tissue, mention section thickness, mounting medium, and coverslip type.
State your time constraints. Even for fixed samples, "I have a 2-hour booking slot" helps the LLM estimate whether a large tiling job will finish in time.
Mention your analysis plan. If you plan co-localisation analysis, the LLM will stress sequential acquisition and Nyquist sampling. If you plan phasor FLIM, it may suggest fewer photons are acceptable.
Tell it what you have already tried. "My image is dim and noisy at 40×" gives the LLM something concrete to troubleshoot.
Iterate. Start with a rough plan, ask the LLM to critique it, then refine.
What the LLM can help with
Before your session — feasibility checks, objective and detector selection, acquisition time estimates, automated workflow design, comparing approaches.
During your session — quick hardware reference, troubleshooting dim or noisy images, sanity-checking unexpected signal (autofluorescence vs bleed-through).
After your session — planning analysis pipelines, writing methods paragraphs, calculating correct scale bars, preparing figures.
What the LLM cannot do
  • It cannot control the microscope or acquire images.
  • It does not know the live state of the hardware (which objectives are on the nosepiece right now, whether the incubator is running).
  • It cannot replace FILM staff training — especially laser safety, alignment, and system-specific procedures.
  • It does not know your sample's actual brightness or labelling efficiency. You must provide these or ask it to look up published values.
  • Quantitative analysis (lifetime fitting, cell counting, deconvolution) requires dedicated software. The LLM can help you choose settings but not run the analysis.
For anything the LLM cannot resolve, contact FILM staff or book an Image Analysis Support session.
Quick-start checklist
Before your LLM conversation, gather the following and paste it into your prompt alongside HCF4.md:
  • Fluorophore(s) and their excitation / emission maxima
  • Sample type (cells, tissue section, organoids, whole mount; fixed or live)
  • Sample thickness (monolayer, cryosection thickness, cleared tissue depth)
  • Substrate (glass-bottom dish, chamber slide, well plate, standard slide; coverslip #1, #1.5, or none)
  • Mounting medium (for fixed) or culture medium (for live)
  • What you want to produce (overview image, 3D reconstruction, colocalisation map, lifetime map, cell count, tiled section)
  • Spatial resolution needed (whole-section overview, cellular detail, subcellular structures)
  • Time resolution needed (single snapshot, Z-stack, time-lapse interval and total duration)
  • Any constraints (booking length, laser safety training status, sample viability)
Useful links
FLIM analysis software: SymPhoTime 64 · NovaFLIM · FLIMlib · FLIMfit
General microscopy education: MicroscopyU · PicoQuant FLIM Applications
LLM tools: Claude · Claude Projects guide