This page holds the method and the per-frame numbers behind Convert to FFF Changes the Raw Data, and Lightroom Renders It Differently. The post has the conclusions, and this page has enough detail to repeat the tests on your own files.
The test files
Five 3FRs from one X2D II, chosen to spread across ISO. Each was converted by Phocus 4.2.3 for Mac: Import, with "RAW File" set to "Convert 3FR to FFF" and "Adjustment" set to "Standard".
| Frame | ISO | Lens | Camera firmware | 3FR size | FFF size | FFF smaller by |
|---|---|---|---|---|---|---|
| B0031524 | 50 | XCD 55V | 1.3.16.1 | 211.0 MB | 112.4 MB | 46.7% |
| B0021315 | 100 | XCD 135 | 1.2.7.16 | 211.2 MB | 133.9 MB | 36.6% |
| B0021134 | 400 | XCD 135 | 1.2.7.16 | 212.4 MB | 151.5 MB | 28.7% |
| B0031107 | 1600 | XCD 55V | 1.3.16.1 | 211.9 MB | 158.9 MB | 25.0% |
| B0021255 | 6400 | XCD 135 | 1.2.7.16 | 212.7 MB | 155.9 MB | 26.7% |
Software: Phocus 4.2.3, Lightroom Classic 16.0, LibRaw 0.22.3, ExifTool, xz 5.8.4, on macOS 27.2.
Raw data: how the two files were compared
LibRaw's unprocessed_raw writes the sensor mosaic out as 16-bit values with no demosaicing, white balance or scaling. I ran it on both files of each pair. The 3FR's raw strip is 11904 x 8842, uncompressed, 16 bits per photosite. The FFF's is the same dimensions, stored as lossless JPEG.
For each of the four positions in the Bayer pattern I fitted a straight line, FFF value against 3FR value, by least squares, over photosites whose 3FR value was between 4,200 and 60,000. In the two pairs where I read them, the black level tag is 4096 and the white level tag 64763 in both the 3FR and the FFF.
| Frame | ISO | Red | Green (both positions) | Blue | Photosites that differ |
|---|---|---|---|---|---|
| B0031524 | 50 | identical | 1.001521x - 6.66 | 1.007810x - 32.47 | 19.5% |
| B0021315 | 100 | identical | 1.001525x - 6.73 | 1.007812x - 32.49 | 52.2% |
| B0021134 | 400 | identical | 1.001528x - 6.79 | 1.007813x - 32.50 | 65.9% |
| B0031107 | 1600 | identical | 1.001526x - 6.75 | 1.007812x - 32.49 | 67.2% |
| B0021255 | 6400 | identical | 1.001525x - 6.74 | 1.007812x - 32.50 | 66.2% |
Red is byte for byte the same in every pair. Every green and blue photosite is within one count of its line, and the residual after the fit has a standard deviation of 0.2 to 0.3 counts. A single line per color fits the whole frame that closely, which leaves no room for a correction that differs across the sensor.
The blue multiplier is 129/128 (1.0078125) to the precision of the fit. Solving each line for the value that maps to itself gives about 4,430 for green and about 4,160 for blue, both close to the black level. Values near black therefore come through unchanged, which is why the dark ISO 50 frame has the fewest changed photosites.
The compression adds no error. The differences are fully accounted for by the scaling and its rounding to whole numbers.
A single pair converted by Phocus 4.2.1 in July 2026 gave the same lines (green 1.001503x - 6.56, blue 1.007807x - 32.45). A tethered FFF from the same camera, checked the same month, had gaps in its histogram at the positions this scaling produces, the same positions as a converted file.
Researching and writing the blog and page posts I produce takes up a lot of my time and costs me money for software subscriptions, tooling, and other incidentals related to running a business.
Please consider a paid membership to help me keep the lights on and pay the bills - it is only a fraction of the cost of an X2D II and a lens. Click on the button to go to the membership page to see the options available.
What else differs in the file
Read with ExifTool. The model, preview and ColorMatrix1 rows were checked on all five pairs, the compression and level rows on two:
| 3FR | FFF (4.2.3 conversion) | |
|---|---|---|
| Model tag | X2D II 100C |
Hasselblad X2D II 100C |
| Raw compression | Uncompressed | JPEG (lossless) |
| Preview in the first image directory | 3888 x 2918 JPEG | 971 x 728, uncompressed |
| Other previews | none | two more at 971 x 728 (one 16-bit) |
| ColorMatrix1 tag | present | absent |
| Black level, white level | 4096, 64763 | 4096, 64763 |
Applying an adjustment to an FFF in Phocus rewrites the FFF file itself. A 3FR is left byte for byte alone and gets a .phos sidecar.
Phocus exports
Each of the ten files had the factory "Standard" adjustment preset applied individually, then was exported with the factory TIFF-16 output preset: 16-bit, uncompressed, Adobe RGB (1998), 11656 x 8742. Applying a preset with several images selected only changed the active image, so each file was done on its own and its preset label checked.
Pixel data was read straight from the single uncompressed strip of each TIFF and compared sample by sample.
| Frame | ISO | Channel averages, FFF / 3FR (R, G, B) | Median absolute difference (R / G / B) | Samples identical |
|---|---|---|---|---|
| B0031524 | 50 | 0.99991, 1.00004, 0.99992 | 21 / 11 / 16 | 6.2% |
| B0021315 | 100 | 1.00000, 1.00000, 1.00000 | 0 / 0 / 0 | 97.9% |
| B0021134 | 400 | 1.00002, 1.00000, 1.00000 | 39 / 29 / 37 | 1.6% |
| B0031107 | 1600 | 1.00005, 1.00000, 1.00001 | 47 / 34 / 37 | 2.9% |
| B0021255 | 6400 | 1.00004, 1.00001, 1.00001 | 61 / 50 / 59 | 0.7% |
Differences are on the 16-bit scale, 0 to 65,535. For scale, the median difference between horizontally adjacent green samples inside a single 3FR export runs from 71 (ISO 50) to 644 (ISO 6400).
Control: one 3FR exported twice produced two byte-identical TIFFs.
Lightroom Classic exports
All ten files were added to a catalog in place, with no develop preset applied at import. Exports were 16-bit TIFF, uncompressed, Adobe RGB (1998), no resizing, no output sharpening. Lightroom embeds its develop settings in the files it exports, so the settings of each pair were read back with ExifTool and compared.
"As Shot" white balance as Lightroom read it:
| Profile | Every 3FR | Every FFF |
|---|---|---|
| "Camera Standard" (applied by default) | 5500 / +10 | 5450 / +11 |
| "Adobe Standard" | 5100 / +22 | 5100 / +23 |
With all ten files set to "Daylight" (5500 / +10), none of the 120 embedded develop settings differed within any pair. Channel averages of the FFF export relative to the 3FR export:
| Frame | ISO | Camera Standard: R | G | B | Adobe Standard: R | G | B |
|---|---|---|---|---|---|---|---|
| B0031524 | 50 | +0.21% | +0.11% | -0.83% | +0.53% | +0.44% | -0.24% |
| B0021315 | 100 | +0.07% | +0.01% | -0.46% | +0.07% | +0.02% | -0.51% |
| B0021134 | 400 | +0.14% | +0.07% | -0.48% | +0.13% | +0.07% | -0.48% |
| B0031107 | 1600 | +0.16% | +0.27% | -0.45% | +0.63% | +0.16% | -0.43% |
| B0021255 | 6400 | +0.12% | +0.08% | -0.33% | +0.12% | +0.06% | -0.35% |
As imported, each file at its own "As Shot":
| Frame | ISO | Camera Standard: R | G | B | Adobe Standard: R | G | B |
|---|---|---|---|---|---|---|---|
| B0031524 | 50 | -0.06% | -0.07% | +0.04% | +0.65% | +0.34% | -0.01% |
| B0021315 | 100 | -0.04% | -0.06% | +0.13% | +0.13% | -0.03% | -0.20% |
| B0021134 | 400 | -0.07% | -0.07% | +0.20% | +0.21% | -0.03% | -0.18% |
| B0031107 | 1600 | -0.04% | -0.07% | +0.19% | +0.68% | +0.02% | -0.17% |
| B0021255 | 6400 | -0.06% | -0.05% | +0.12% | +0.20% | -0.01% | -0.15% |
The shift by region and by brightness
For the ISO 400 frame under "Camera Standard" at "Daylight", I split the frame into a 6 x 8 grid and into brightness bands. The FFF had more red and less blue in all 48 regions. By brightness of the 3FR export:
| Brightness band (16-bit) | Share of frame | Red | Green | Blue |
|---|---|---|---|---|
| 3,000 to 8,000 | 27.0% | +0.20% | +0.06% | -1.10% |
| 8,000 to 16,000 | 16.7% | +0.18% | +0.12% | -0.76% |
| 16,000 to 30,000 | 19.2% | +0.16% | +0.11% | -0.62% |
| 30,000 to 45,000 | 12.9% | +0.15% | +0.09% | -0.46% |
| 45,000 and up | 23.6% | +0.10% | +0.05% | -0.33% |
The shift is larger in the shadows than in the highlights.
Have you seen the guide? I've published Essential Phocus 4.x for Mac - 85 topics across 8 sections and 251 pages covering everything from HNCS color science to HDR workflows. It's the reference manual Hasselblad hasn't updated since 3.8. It's $49, and updates are included.
Repeatability
Exporting the same pair twice in one session gave pixel-identical TIFFs.
The ISO 400 pair was also exported a second time on a separate occasion, with the same settings recorded in the files. Those two runs were not pixel-identical: about half the samples matched, the 95th percentile difference was 50 and the largest 142 on the 16-bit scale, and the channel averages of both files were about 0.03% higher in the second run. The difference between the 3FR and the FFF was the same in both, +0.13% red, +0.07% green and -0.48% blue. I do not know what changed between the runs.
In the second run the per-sample differences, FFF against 3FR, were 40 / 103 / 204 for red, 23 / 77 / 167 for green and 107 / 190 / 253 for blue, as 50th / 95th / 99th percentiles.
The FFFs Lightroom was given
The FFFs Lightroom saw in the five-frame runs had already been opened in Phocus and had "Standard" applied, which rewrites the FFF file. An earlier run on a single pair used an FFF converted by Phocus 4.2.1 in July and untouched since. It gave the same "As Shot" reading, 5450 / +11, and with settings matched it shifted the same way (in ProPhoto RGB: red 0.00%, green +0.28%, blue -0.45%).
The camera profile digest embedded in the exports matched within every pair. Process Version was 15.4 throughout.
Lightroom "Auto"
"Auto" pressed by hand in the Basic panel on the ISO 400 pair, Lightroom Classic 16.0, "Camera Standard":
| Slider | 3FR, "As Shot" | FFF, "As Shot" | 3FR, "Daylight" | FFF, "Daylight" |
|---|---|---|---|---|
| Exposure | +0.70 | +0.71 | +0.70 | +0.69 |
| Contrast | +6 | +6 | +6 | +6 |
| Highlights | -62 | -62 | -62 | -60 |
| Shadows | +72 | +72 | +72 | +72 |
| Whites | +36 | +41 | +36 | +36 |
| Blacks | -33 | -35 | -33 | -32 |
| Vibrance | +20 | +20 | +20 | +20 |
| Saturation | +4 | +4 | +4 | +4 |
Pressing "Auto" a second time returned the same values.
The blind test
From the ISO 400 "Camera Standard" exports I cut the same two 2400 x 1600 regions, one of water and one of sky, at "As Shot" and at "Daylight". The crops kept their 16-bit depth and Adobe RGB profile. Ten pairs were given coded names in a random order: six paired a 3FR render with an FFF render, four paired a file with itself. Each pair was viewed in Quick Look, flicking between the two with the arrow keys.
Color difference in each crop pair, as Delta E 2000 computed on 16 x 16 pixel block averages (50th percentile / 95th percentile / largest block):
| State | Crop | Delta E 2000 | Measured direction |
|---|---|---|---|
| "As Shot" | water | 0.20 / 0.26 / 0.46 | FFF slightly cooler |
| "As Shot" | sky | 0.14 / 0.21 / 0.35 | FFF slightly cooler |
| "Daylight" | water | 0.40 / 0.46 / 0.57 | FFF slightly warmer |
| "Daylight" | sky | 0.28 / 0.42 / 0.50 | FFF slightly warmer |
The ten trials:
| Trial | Pair | Answer | Scored |
|---|---|---|---|
| 1 | "As Shot" water, same file twice | no difference | correct |
| 2 | "As Shot" sky, FFF then 3FR | second possibly a touch warmer | matches the measured direction |
| 3 | "As Shot" sky, same file twice | no color difference | correct |
| 4 | "As Shot" water, FFF then 3FR | can't tell | not seen |
| 5 | "Daylight" sky, same file twice | can't tell | correct |
| 6 | "Daylight" water, same file twice | can't tell | correct |
| 7 | "Daylight" water, FFF then 3FR | can't tell | not seen |
| 8 | "As Shot" water, 3FR then FFF | can't tell | not seen |
| 9 | "Daylight" water, 3FR then FFF | second looks a little warmer | matches the measured direction |
| 10 | "Daylight" sky, 3FR then FFF | can't see a difference | not seen |
Two of six real pairs were called, both in the direction the measurements give. None of the four identical pairs was called as a color difference. Six real pairs is too few to turn into a detection rate.
Compression
Each 3FR was piped through four compressors and the output size recorded. Sizes in MB, with the saving against the original:
| Frame | ISO | 3FR | gzip -9 | zstd -19 | xz -6 | xz -6 with delta filter | FFF |
|---|---|---|---|---|---|---|---|
| B0031524 | 50 | 211.0 | 131.8 (37.5%) | 113.6 (46.2%) | 95.9 (54.5%) | 93.8 (55.6%) | 112.4 |
| B0021315 | 100 | 211.2 | 172.8 (18.2%) | 165.4 (21.7%) | 140.9 (33.3%) | 126.4 (40.2%) | 133.9 |
| B0021134 | 400 | 212.4 | 178.3 (16.1%) | 171.2 (19.4%) | 157.0 (26.1%) | 147.0 (30.8%) | 151.5 |
| B0031107 | 1600 | 211.9 | 180.7 (14.7%) | 177.7 (16.1%) | 157.4 (25.7%) | 151.1 (28.7%) | 158.9 |
| B0021255 | 6400 | 212.7 | 180.6 (15.1%) | 175.4 (17.5%) | 158.8 (25.4%) | 149.8 (29.6%) | 155.9 |
gzip -9 uses deflate, the same method as an ordinary ZIP file. The delta filter is --delta=dist=4: a raw value is two bytes, and the same color repeats every second photosite along a row, so the same-color neighbor is four bytes back.
On an M4 Pro Mac mini using all cores (-T0), xz with the delta filter took 12 to 17 seconds per file, and restoring one took 0.7 seconds. A file compressed with xz -k -T0 --delta=dist=4 --lzma2=preset=6 and restored with xz -dk matched the original byte for byte.
On two of the files I also tried faster settings: zstd at levels 1 to 3 saved 13% to 18%, and lz4 saved 1% to 4%.
Limits
- One camera body. Whether the multipliers are the same on another X2D II is not known.
- One frame per ISO, each of a different scene, so ISO and scene content cannot be separated. That matters most for the ISO 100 pair, which rendered 97.9% identical in Phocus where the others did not.
- Lightroom Classic only, with two profiles. Capture One was not tested.
- The blind test is one person, one frame, ten pairs.
- Why Lightroom's two renders differ, and why they differ in the direction they do, is not established by any of this.