Tutorial: a time-lapse of two H-alpha captures¶

This walks through a real pair of captures from start to finished movie: two H-alpha videos of the whole disc taken about five minutes apart, 15_32_09.avi and 15_37_19.avi (720 by 540 pixels, 8-bit, 1000 frames each). Two frames is the smallest time-lapse there is, but it shows every part of the method; with more captures you add them to the list. You need Helia 0.6 or later and, for an MP4, ffmpeg. The method is described in Time-lapse.

1. What the two captures look like¶

Stacked one by one (helia stack 15_32_09.avi -o a.png --preset ha --colour auto) the two discs are 17 pixels apart across and 6 down: the telescope drifted between the captures. The second is also about 13% dimmer over the disc. In an animation made from the plain stacks the Sun would jump and flicker.

An AVI has no timestamps, so Helia reads the time from the file name: 15_32_09 is 15:32:09. The clock of the capture program was British Summer Time, an hour ahead of UTC, and the date is not in the name, so the command line says both: --date 2026-10-20 --utc-offset 1. That makes the frames 14:32:09 and 14:37:19 UTC, 5 minutes and 10 seconds apart.

2. Make the time-lapse¶

On the command line, with the stack settings after -- (the filter, as for any stack of these captures):

helia timelapse 15_32_09.avi 15_37_19.avi -o tl \
    --date 2026-10-20 --utc-offset 1 --crop --deflicker match \
    --movie auto --fps 2 -- --preset ha

-o tl is a directory. --crop cuts every frame to the area both cover, --deflicker match brings the two to the same brightness and --fps 2 makes the movie two frames a second. Everything after -- is what you would give helia stack; here the filter is the only thing, as in the prominence tutorial.

What you should see (the two stacks take a few seconds each on a GPU workstation):

[1/2] stacking 15_32_09.avi
[2/2] stacking 15_37_19.avi
registering 2 frames
2 frames registered to frame 2 (1 limb+phase, 0 limb, 0 phase, 0 failed); largest shift 18.2 px
cropped to the common area: 698 x 530 at (2, 8)
brightness gains 0.937 to 1.073
    1 14:32:09  shift ( 17.17,  -5.94) px  limb+phase  phase PSR 180
    2 14:37:19  shift (  0.00,   0.00) px  reference
wrote 2 frames to tl (tl/timelapse.csv)
wrote tl/timelapse.mp4 (mp4)

With two frames the middle frame, the default reference, is the second, so the first is the one that moves. In the app, open the first capture, choose H-alpha in the Open step, then go to the last step, Time-lapse, press Add captures and choose both files (or Add the open capture and then add the second). Type the date and the UTC offset in Times, tick Crop to the common area, choose Match under Brightness, and press Make time-lapse of 2.

The Time-lapse step on these captures. The line under the picture says frame 1 was moved by (+17.2, -5.9) pixels, by the limb fit and the phase correlation together, and that the correlation's peak was 181 times the level of its surroundings. The registered disc does not move when you scrub to frame 2.

The stacks in the app use the Stack step's settings, and the app, like helia stack, moves off a clipped colour plane by itself. These H-alpha captures clip in red (92 to 100% of the lit pixels sit at the sensor limit), so when you open one the Stack step says "Channel: the red plane is clipped ... stacking the green plane instead" and the Channel chip under Advanced settings shows G. You do not have to do anything; choose a channel yourself on that chip only to stack another plane (a plane you pick is never changed).

3. Read what it did¶

The table tl/timelapse.csv has the numbers behind each line:

frame,file,source,time_utc,shift_x_px,shift_y_px,method,limb_x,limb_y,limb_radius,limb_confidence,phase_dx,phase_dy,phase_psr,level,gain,derotated_px
1,frame-0001.png,0001-15_32_09.tif,2026-10-20T14:32:09.000,17.175,-5.935,limb+phase,288.473,289.039,222.359,0.789,0.179,-0.401,180.276,0.64471,0.9365,
2,frame-0002.png,0002-15_37_19.tif,2026-10-20T14:37:19.000,0.000,0.000,reference,271.477,294.573,221.957,1.000,,,,0.56287,1.0727,

• The limb fits put the disc centres at (288.47, 289.04) and (271.48, 294.57): the limb alone says the first frame is (17.00, -5.53) pixels from the second.
• The phase correlation of the middle of the disc then adds (0.18, -0.40) pixels, with a peak-to-sidelobe ratio of 180 (a solid lock is above 8; Helia uses it above 6). It follows the granulation and filaments, so it corrects what an edge the seeing has moved cannot say. The shift applied is the sum, (17.17, -5.94).
• The radius differs by 0.4 pixel between the two (222.36 and 221.96): the disc's size as the limb fit sees it varies a little with seeing and focus. Helia moves the picture but does not rescale it.
• level is the mean brightness of the inner disc: 0.645 and 0.563. gain brings both to their median, 0.604, by 0.937 and 1.073.

Mean brightness of the inner disc of each frame in timelapse.csv, and the median both are brought to by the gain.

To see whether the registration worked, subtract the two frames. Before registration the two stacks differ by a bright crescent where the limbs do not coincide, and the spots, filaments and plage are doubled. After it the difference is a faint, smooth pattern: what really changed in five minutes, and the brightness differences that a single gain cannot take out.

Left: the difference of the two stacks as stacked. Right: the difference of frames 1 and 2 after registration, on the same scale. Inside 0.9 of the radius the mean absolute difference falls from 0.064 to 0.026 and the RMS from 0.071 to 0.036 of full scale.

The animated PNG that Helia writes when there is no ffmpeg is the same movie. This is the one from these two frames at two frames a second; in a browser it should show a disc that stays put while the grain and the filaments shift.

4. The files¶

tl/ now holds:

frame-0001.png  frame-0002.png   the registered, cropped frames (698 x 530)
timelapse.mp4                    H.264, 2 frames a second (timelapse.png instead without ffmpeg)
timelapse.csv                    the table above
stacks/0001-15_32_09.tif  stacks/0002-15_37_19.tif   the stacks, as stacked

The MP4 was checked here with ffmpeg 7.0.2: H.264, yuv420p, 698 by 530, 2 frames a second, one second long.

To try other registration settings without stacking again, add --reuse-stacks (the stacks in tl/stacks are used as they are):

helia timelapse 15_32_09.avi 15_37_19.avi -o tl --date 2026-10-20 --utc-offset 1 \
    --reuse-stacks --no-refine --reference first --format tiff --movie none

--no-refine registers by the limb alone, --reference first keeps the first frame where it is (so the second moves, and more of the frame is cropped), --format tiff writes 16-bit TIFF frames and --movie none no movie.

To add more frames, list more captures. They are sorted by their times, so the shell's alphabetical order is fine for SharpCap names. Captures that are already stacked images can be listed too.

What to expect and what can go wrong¶

For a time-lapse over hours, the Sun's surface features also turn with the Sun's rotation, about 13 degrees a day; over these five minutes that is 0.05 degree, about a fifth of a pixel at the limb of a disc 222 pixels in radius, so nothing was done about it here.